# Zero Day Playbook: every page as markdown A reference that lists and does not rank. Runbooks for the first 72 hours after a critical disclosure, by role and asset type, plus the Disclosure Clock of 37 headline bugs from 2026. --- # Zero day playbook: a 72-hour checklist for your role and the asset that was hit PLAYBOOK, read 19 Nov 2026 **The short answer.** Zero Day Playbook turns an advisory into a checklist. Pick your role and the asset that was hit, set the hour, and see which 72-hour checklist steps are open or done in the 0 to 4, 4 to 24 and 24 to 72 hour windows. The Disclosure Clock behind it: 33 bugs with a dated exploitation, median gap 2 days. ## The 72-hour clock Hour 4 of 72, the 0 to 4 hours window is open. Executive facing a network edge appliance (VPN or firewall): 8 of 19 steps open or done. ### 0 to 4 hours 1. [ ] Executive: Fix hour zero in writing: the earlier of the time the vendor published the advisory and the time your team learned of it, with its time zone. Name the one person who can approve downtime and the one who can accept risk, and set the bridge cadence: every 4 hours until the security lead reports the fix verified on every in-scope asset, then daily. 2. [ ] Executive: Get the five-line brief from the security lead and use the same five lines for every later update. A filled example for the Palo Alto PAN-OS advisory of 14 Oct 2026, CVE-2026-0322 (the company figures are invented): Affected: 6 firewalls, 4 with the management interface reachable from the internet. Exploitation: confirmed by Palo Alto and listed by CISA on 14 Oct; no sign of access on our devices yet, log search unfinished. Done: management interfaces limited to the jump box. Open: 6 firewalls unpatched, Palo Alto lists no workaround and the jump box only limits who can reach the flaw, owner named, compromise search due 15 Oct 12:00. Next update: 15 Oct 08:00. 3. [ ] Executive: Approve emergency change authority in writing, with limits: the actions allowed without the change board (apply a mitigation, apply the fixed build, restart the service), the systems covered, the people who may act, and an expiry 72 hours after hour zero. Removal of a safeguard stays with the normal board. 4. [ ] Security lead: List every appliance of the vendor, its model family and software version. PAN-OS (CVE-2026-0322): run show system info and read model and sw-version, or read Dashboard > General Information; check Panorama and each firewall it manages. SonicOS (CVE-2026-19868): read the firmware version in the management interface for every firewall of a pair and every NSv. Match each version to the vendor's fixed-build table for its own release train. 5. [ ] Security lead: Check the exposure conditions the advisory names. PAN-OS CVE-2026-0322: read Device > Setup > Interfaces > Management Interface Settings > Permitted IP Addresses, and Network > Network Profiles > Interface Mgmt for any profile that allows HTTPS or SSH on a data-plane interface. Then test from a host outside your network, such as a cloud instance, with nmap -Pn -p 22,443 and curl -sk -m 5 -o /dev/null -w "%{http_code}\n" https:///; any answer means the interface is reachable from the internet. SonicOS (CVE-2026-19868) has two preconditions. First, SAML is the login method for SSLVPN users: review the SSLVPN user authentication settings in the management interface. Second, at least one LDAP, RADIUS or TACACS+ server is active: open Device > Users > Settings > Authentication, then Configure RADIUS > Settings > RADIUS Servers, Configure LDAP > Settings > LDAP Servers and Configure TACACS+ > Settings > TACACS+ Servers; any server listed and not disabled counts. If either half fails, the bypass does not apply to that firewall. The attacker also needs valid credentials for one of those alternate methods, so list who holds accounts in those directories and confirm whether the SSLVPN portal answers from an untrusted network. 6. [ ] Platform engineering: Cut reachability before the patch. For PAN-OS, set Permitted IP Addresses to one jump box, which is the only system allowed to reach the management interface. For SSLVPN flaws, restrict the portal to known source ranges only if remote staff can work without it. Test from an untrusted network that the management address no longer answers. 7. [ ] Security lead: Pull forensic evidence before any upgrade. On PAN-OS export device state (Device > Setup > Operations > Export device state) and save System, Config and Authentication logs from Monitor > Logs. Save the configuration audit and the administrator list (Device > Administrators). Send logs to storage the appliance cannot write to. 8. [ ] Security lead: Search the exported logs and the firewall traffic logs for the indicator addresses the vendor published. For CVE-2026-0322 these are 66[.]42[.]72[.]14, 208[.]83[.]239[.]114 and 149[.]248[.]47[.]234. Palo Alto published no file or log signatures, so also list every administrator login to the management interface in the exposure window and check each against a person. ### 4 to 24 hours 9. [ ] Executive: Decide on downtime with three facts from the security lead: whether exploitation is confirmed, whether an unauthenticated network path exists, and whether a mitigation limits the reachable set to trusted networks. Ask the business owner for two numbers: cost per hour offline and the manual process that runs meanwhile. Then choose one: keep running behind the mitigation, run a reduced service, or take it offline until the fix is verified. 10. [ ] Executive: Ask counsel two questions in writing: which notification clocks apply to us if a compromise is found, and who makes the materiality determination if we are a US registrant. SEC Form 8-K Item 1.05 is due "within four business days after the registrant determines that it has experienced a material cybersecurity incident," and the determination "must be made without unreasonable delay after discovery of the incident." Put the determination meeting in the calendar with a time. The deadlines for each regime are on the communications playbook. 11. [ ] Executive: Send the board brief 24 hours after hour zero, sooner if exploitation is confirmed on your own systems. It holds the five lines, the decisions taken with the time of each, the decisions that need the board, and what is unknown. Use dates and times, not words like recently. 12. [ ] Platform engineering: Upgrade to the fixed build for the train, in a window within the day. PAN-OS (CVE-2026-0322), by release train: 12.2 to 12.2.4; 12.1 to 12.1.4-h12, 12.1.7-h8 or 12.1.10-h2; 11.2 to 11.2.4-h22, 11.2.7-h21, 11.2.10-h15, 11.2.13-h3 or 11.2.16; 11.1 to 11.1.4-h37, 11.1.6-h39, 11.1.7-h11, 11.1.10-h35, 11.1.13-h15 or 11.1.16-h3; 10.2 to 10.2.7-h38, 10.2.10-h41, 10.2.13-h26, 10.2.16-h11 or 10.2.18-h12 (each or later in its sub-train, per Palo Alto). SonicOS (CVE-2026-19868): Gen7 to 7.3.4-7013 or higher, Gen8 to 8.2.3-8015 or higher. In an HA pair, suspend the passive node, upgrade and verify it, fail over, then upgrade the other, and take a configuration backup first. On PAN-OS run show system info after each node and confirm sw-version equals the fixed build. 13. [ ] Platform engineering: For devices you cannot upgrade today, keep the reachability cut in place and apply the vendor's partial coverage if it exists. PAN-OS (CVE-2026-0322): Threat IDs 510046 and 510047 (content version 9159-10287 or later) need management traffic routed through a data-plane interface with decryption and threat prevention; treat them as partial cover only. SonicOS (CVE-2026-19868): the workaround is to disable the LDAP, RADIUS and TACACS+ servers, which removes the alternate login method the bypass needs; use it only where no one needs those servers. ### 24 to 72 hours 14. [ ] Executive: Review each risk acceptance as a signed record. Template, copy and fill: Asset: ____. Exposure (who can reach it, from where): ____. Flaw (CVE, authentication needed, impact): ____. Compensating control in place and tested on: ____. Reason it cannot be fixed now: ____. Owner: ____. Approver: ____. Expiry (date and hour): ____. Reject any acceptance for an unauthenticated flaw with total control and no compensating control. Set the expiry of an internet-reachable asset at 7 days or less, since 24 of the 33 dated bugs in the Disclosure Clock were exploited within that time. 15. [ ] Executive: Ask for evidence, not status: the version output from the running service on each asset, and the security lead's triage outcome, one of CISA's three: no evidence of compromise, compromise suspected, compromise confirmed. Do not accept a change ticket as proof of a fix. If the outcome is suspected or confirmed, move the incident to the full incident response plan and call counsel the same hour. 16. [ ] Executive: Call the 72-hour review and assign each action item an owner and a date. Decide what to pre-approve for next time (emergency authority, a standing downtime rule), which end-of-life products to retire, and what to budget. Close with the hour-zero to verified-fix time as the number the board sees. 17. [ ] Security lead: PAN-OS (CVE-2026-0322): hunt for persistence on every device that was reachable before the cut. Open Device > Config Audit and compare the running configuration with the last backup taken before the device was first reachable from an untrusted network, not the 14 Oct 2026 disclosure date, because exploitation can predate an advisory; with no such backup, use the oldest one you hold and treat each difference as unexplained until a person owns it. Open Device > Administrators and match every account and role to a person. Read Monitor > Logs > Configuration and System for administrator logins, and for changes made by accounts nobody can explain. Check Device > Certificate Management > Certificates for certificates you did not install, and Device > Server Profiles and Device > Authentication Profile for changed RADIUS, LDAP or SAML servers. Check Network > Virtual Routers and Policies for routes, NAT and rules added since that last clean backup. Because the flaw runs code as root, a device with any unexplained change is rebuilt, not cleaned. SonicOS (CVE-2026-19868): compare SSLVPN login records with the SAML provider's sign-in log: a user on the firewall with no matching SAML sign-in came in through the alternate method. SonicWall names no log event for this bypass, so export the SSLVPN login events from the firewall event log or the syslog server for the exposure window into firewall-sslvpn-users.txt (one username per line) and the SAML sign-ins for the same window from the identity provider into idp-saml-signins.txt, then run comm -23 <(sort -u firewall-sslvpn-users.txt) <(sort -u idp-saml-signins.txt). Every name it prints logged into SSLVPN without a SAML sign-in and gets an owner's explanation, or its credentials are rotated and its sessions ended. 18. [ ] Platform engineering: If any device was reachable and cannot be cleared, rotate what it holds: administrator and service passwords, LDAP and RADIUS bind credentials, VPN pre-shared keys, SAML signing and device certificates, and force all VPN sessions to end. 19. [ ] Security lead: Rescan from outside the network for the vulnerable version banner and the management port, and confirm each device meets its fixed build. Close the temporary jump-box rule only when the owner agrees, and keep it if no one needs the management interface from elsewhere. ## What the checklist is built from, and how it changes The checklist merges the role playbook with the asset playbook you pick, starting on the executive and network edge appliance playbooks. Each vendor step rests on the advisory for the CVE it names. The board brief example uses invented company figures. The sources are listed below, each with the day it was read, 19 Nov 2026. The hour slider opens a window once the hours elapsed pass the end of the one before: at hour 4 only the 0 to 4 hours window is open, and the 4 to 24 hours window shows from hour 5. Role and asset choices change the steps, commands and fixed builds; there are 17 playbooks to choose from. ## When the vendor has no fixed build Some advisories list no fixed build for your release line, or none yet. Then the steps above that need a patch cannot close the work, and three things take their place. Isolate: take the device off the internet when its interface is reachable from untrusted networks and exploitation is reported. Compensate: keep the reachability cut, apply any partial vendor cover, and test both from outside. Set the review clock: the executive names one owner and a review time, 24 hours at most, and reviews daily once exploitation is confirmed. Record the three in a signed risk acceptance, step 14, and move to the fixed build the day the vendor ships it. ## A worked case: SonicOS CVE-2026-19868 One real case, SonicOS CVE-2026-19868, hour by hour from the dated events in the sources. Hour 0 is 00:00 UTC on 5 Nov 2026, the start of the day SonicWall dated advisory SNWLID-2026-0014 version 1.0 (the revision history gives a date, not a time; the Disclosure Clock uses the same date). SonicWall gives no date for the first attack and no indicators. The Disclosure Clock dates first exploitation at 6 Nov 2026, the day NVD published the record and CISA listed the flaw as exploited, so that date is a catalog date, not an attack date. Three events fall inside the 72 hours and one after it: the 9 Nov revision, between hour 96 and hour 120 because it carries only a date. - Hour 0 to 24, 5 Nov: version 1.0 is released with Gen7 fixed at 7.3.4-7013 and Gen8 at 8.2.3-8015. The team works steps 1, 4 and 5. - Hour 34, 10:40 UTC on 6 Nov: the PSIRT data stamps the advisory as published. The team works steps 6 and 7. - Hour 43, 19:17 UTC on 6 Nov: NVD publishes the record, and CISA adds the flaw to its catalog the same day (listing time not given, so between hour 24 and hour 48) with a due date of 20 Nov and exploitation marked active. The team works steps 12 and 13 and rescans from outside before hour 72. - Hour 96 to 120, 9 Nov: version 1.1 gives the workaround as disabling every LDAP, RADIUS and TACACS+ server, which stops anything that signs in through them. The team, past its 72-hour report, rechecks each device still on the workaround and works steps 17 and 18. ## Every playbook, by role and by asset type - [Communications](/playbooks/communications/) - [Executive](/playbooks/executive/) - [Endpoint fleet](/playbooks/endpoint-fleet/) - [Dependency](/playbooks/dependency/) - [Network edge appliance (VPN or firewall)](/playbooks/network-edge-appliance/) - [Cisco ISE network access control](/playbooks/network-access-control/) - [Container base image](/playbooks/container-base-image/) - [SaaS vendor](/playbooks/saas-vendor/) - [Email gateway](/playbooks/email-gateway/) - [Remote support tool](/playbooks/remote-support-tool/) - [Artifact repository](/playbooks/artifact-repository/) - [CMS and storefront](/playbooks/cms-and-storefront/) - [Windows operating system](/playbooks/windows-endpoint-os/) - [Linux server (SSH)](/playbooks/linux-server-ssh/) - [AI infrastructure](/playbooks/ai-infrastructure/) - [Developer collaboration platform](/playbooks/developer-collaboration-platform/) - [Application runtime](/playbooks/application-runtime/) ## The Disclosure Clock in the same 72 cells Of the 33 headline bugs of 2026 with a dated first exploitation, 12 were exploited on or before disclosure day, 16 by hour 24, 18 by hour 48 and 19 by hour 72. The median gap is 2 days. One row per dated bug, CC BY 4.0: [clock.csv](/datasets/clock.csv), [clock.json](/datasets/clock.json). ## Sources - CISA BOD 26-04, Prioritizing Security Updates Based on Risk: https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk (read 19 Nov 2026) - CISA BOD 26-04 Implementation Guidance, forensic triage steps: https://www.cisa.gov/news-events/directives/bod-26-04-implementation-guidance-prioritizing-security-updates-based-risk (read 19 Nov 2026) - SEC final rule, Cybersecurity Risk Management, Strategy, Governance, and Incident Disclosure, 88 FR 51896 (4 Aug 2023), Form 8-K Item 1.05: https://www.federalregister.gov/documents/full_text/html/2023/08/04/2023-16194.html (read 19 Nov 2026) - Palo Alto Networks advisory CVE-2026-0322: https://security.paloaltonetworks.com/CVE-2026-0322 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-0322: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322 (read 19 Nov 2026) - NVD record for CVE-2026-0322: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-0322 (read 19 Nov 2026) - SonicWall advisory SNWLID-2026-0014 (CVE-2026-19868): https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-19868: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19868 (read 19 Nov 2026) - The Disclosure Clock data file, zerodayplaybook.com (33 bugs with a dated first exploitation): https://zerodayplaybook.com/datasets/clock.csv (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/ --- # A zero day response plan is ten clauses long, and its drill proves that the authority table works Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** A zero day response plan has 10 clauses: scope, triggers, roles, standing approvals, inputs, deadlines, communications, exceptions, records and a drill. The template below fills each one, adds an authority table, and ends with a timed Cisco ISE tabletop that tests approval, inventory and mitigation before a real advisory arrives. ## The plan is the emergency branch of the vulnerability plan, and it settles who may act NIST SP 800-40 Rev. 4, section 3.3, tells organizations to define risk response scenarios in advance: routine patching, emergency patching, emergency mitigation and unpatchable assets. Section 3.5.3 gives the reason: "Without the processes, procedures, and tools in place to implement mitigations, too much time may be lost, and vulnerable assets may be compromised." Section 3.5.2 allows a canary period of "a few minutes to a few hours" even in an emergency. NIST SP 800-61 Rev. 3 separates incident response plans from vulnerability management plans and business continuity plans. The zero day plan belongs to the vulnerability side and hands over to the incident response plan once compromise is suspected, so it names that hand-over. SP 800-61r3 also suggests: "Consider designating an incident lead for each incident." The deadlines in the plan come from how fast attacks start. In the Disclosure Clock, 33 bugs have a dated first exploitation: 16 within one day of disclosure, 24 within seven days, and a median gap of 2 days. By the end of day 3, 19 of the 33 had been exploited, so a plan that takes three days to find an approver acts after more than half of them have been under attack. The role playbooks hold the step-by-step work; the plan holds the authority, the triggers and the deadlines. ## Ten clauses make the plan complete Each row is a clause. Fill the second column with your own facts and name the owner in the third. The fifth column is the record the clause leaves, which a reviewer can ask for. | Clause | What to write | Owner | How it typically fails | Record it produces | | --- | --- | --- | --- | --- | | 1. Scope | What the plan covers: vulnerabilities with evidence of exploitation, and critical flaws in products reachable from outside. The products and asset classes in scope, what is out of scope and who covers it (vendor-hosted services, for example). Review date | Security lead | Vendor-hosted and supplier-run copies are left out, so the first one found mid-event has no owner. | Signed scope list with its review date | | 2. Triggers | The levels in the trigger table below, with who declares each | Security lead | Triggers read only CVSS, so a low-rated bug in the catalog starts nothing. | Trigger table with a version number; each declaration logs the trigger and its source | | 3. Roles and deputies | Event lead, platform lead, application owners, change approver, communications lead, executive approver and scribe. Each has a named deputy and a phone number that works at night | Security lead | A deputy's phone number is out of date, or the deputy has never run an event. | Roster with names, deputies and numbers, and the date of the last call test | | 4. Standing pre-approvals | Mitigations the event lead may apply without the change board, each with its known cost and the people told: restrict a management interface to a jump host, disable a named feature, turn on a WAF rule in block mode, take a service off the internet | Change approver | A pre-approval nobody has costed is vetoed by the service owner on the night. | Pre-approval list with cost, people told and review date | | 5. Inputs | Advisory feeds for every vendor in the inventory, the CISA catalog, and the vendor portal accounts needed to read advisories behind a login, with a named holder and deputy. The inventory fields the first hour needs: product, version, role, exposure, owner, maintenance group | Platform lead | The one holder of a vendor portal login is away, and the advisory sits unread behind it. | Feed and portal-holder list; inventory export with its date | | 6. Deadlines | Event declared within 1 hour of a trigger. Count of exposed assets within 2 hours. A first control on every internet-facing exposed asset within 4 hours. Patch ring 1 within 24 hours. Fix verified or risk accepted in writing within 72 hours. Replace these with figures you can meet | Event lead | The deadlines were copied from a template and never met, so the team ignores them. | Timestamps in the decision log for declaration, count, first control, each patch ring and verification | | 7. Communications | Update interval (4 hours while exploitation is active), the channel, who may speak to customers, regulators and the board, and each notification duty with its trigger and deadline. Leave a duty blank rather than guess | Communications lead | Nobody owns the decision that the organization is aware of an incident, so a notification clock starts unnoticed. | Update log with time, audience and approver; register of notification duties | | 8. Exceptions | A named person and a written answer for each: advisory behind a login, product owned by another team or a supplier, version with no fix, restart refused in business hours, approver away at a weekend, vendor-hosted copy | Security lead | The exception names a team, not a person, and the question lands in a queue. | Exception register: person, written answer, date | | 9. Records | One decision log. Each entry holds time in UTC, decision, owner, reason and expiry. The event closes when every in-scope asset has a verified fixed build or a signed acceptance with an expiry | Scribe | Decisions stay in chat and the log is rebuilt from memory after the event. | The decision log, and its closure entry | | 10. Test and review | The drill below twice a year, a review within 72 hours of every real event, and a version history showing what each review changed | Security lead | The drill is skipped after a reorganization and the roles list names people who have left. | Drill report with times and the plan lines changed; version history | ## Two levels and a compromise flag decide when the plan starts Name the trigger before the day it is needed, so that nobody debates it during the first hour. When sources disagree, the higher level applies. An empty catalog is not a reason to wait. Apache Tomcat CVE-2026-34486 was disclosed on 9 Apr 2026, SOCRadar dates attacks against it from 24 Apr, and CISA listed it on 4 Aug, 102 days after the first attack. | Level | Trigger (any one) | Declared by | Within | What changes | | --- | --- | --- | --- | --- | | Watch | A critical rating (CVSS 9.0 or higher) on a product in the inventory that is reachable from outside, with no statement of exploitation | Analyst on duty | The same business day | Inventory count and mitigation options prepared. No out-of-band changes yet | | Event | The vendor says exploitation is under way; the CVE is in the CISA catalog; a named security firm reports attacks on a product in the inventory | Event lead or deputy | 1 hour from first sight | Pre-approvals are active, the update clock starts, the decision log opens | | Compromise | Evidence of exploitation on one of your own assets | Event lead | At once | Hand over to the incident response plan. Preserve evidence before patching | ## The authority table says who decides what, and for how long The limits are defaults to replace with your own. Their purpose is that every decision on the first night has an owner with a deputy and a ceiling. | Decision | Decides | If unavailable | Limit (default) | Logged as | | --- | --- | --- | --- | --- | | Declare a Watch or an Event | Event lead | Named SOC lead | None | Trigger, source, time | | Apply a pre-approved mitigation | Event lead | Platform lead | Pre-approval list only; reviewed at each update | Mitigation, cost, owner, expiry | | Approve an out-of-band change | Change approver | Deputy approver, then the executive on call | 60 minutes to answer | Request time, answer time, approver | | Take a service offline | Event lead, service owner informed | Executive | 24 hours without owner consent | Service, reason, restore condition | | Accept a risk | Executive | Deputy executive | 30 days; renewal needs a new signature | Asset group, reason, expiry | | Speak to customers or regulators | Communications lead with legal | Executive | Only facts the event lead has confirmed | Statement and approver | | Close the event | Event lead | Security lead | Verified build or signed acceptance for every asset | Closure criteria met | ## When the usual answer is wrong A Watch trigger set at CVSS 9.0 never opens for a bug rated low that is already exploited. Broadcom rates ESX CVE-2026-41709 at CVSS 2.7, and CISA added it to the catalog on 28 Oct 2026. Microsoft's Windows CVE-2026-68820 is rated 7.0 and was marked exploited on the day its fix shipped, 11 Aug 2026. The Event trigger must therefore key on exploitation evidence and catalog listing for products in the inventory, whatever the rating. A one-hour declaration clock assumes the attack starts at disclosure. For BeyondTrust CVE-2026-102764 it started on 29 Aug 2026, 47 days before the 15 Oct advisory, so an Event declared on 15 Oct opens with the Compromise flag already in question and a search window of 47 days. ## The drill: a Cisco ISE tabletop with timed injects Run it twice a year for one hour with the people named in the authority table, using a real advisory. This one is Cisco's cisco-sa-ISE-ABP-VNSW7Tn5, first published 16 Sep 2026 at 16:00 GMT. It is CVE-2026-76460, an authentication bypass in an ISE API, base rating 10.0, and it affects ISE and ISE-PIC regardless of configuration. Cisco lists no workaround, recommends infrastructure access control lists, and names these fixed patches: ISE 3.1 Patch 12, 3.2 Patch 11, 3.3 Patch 12, 3.4 Patch 7 and 3.5 Patch 4. ISE 3.0 is past software maintenance and has no fix. Teams answer from their real inventory. A team that runs no ISE uses an invented card of its own, for example 2 nodes on 3.2 Patch 9, 3 on 3.4 Patch 6 and 1 on 3.0. The scribe records the clock time of every answer. Afterward, compare five times with clause 6: declaration, first inventory answer, first control, approver reached and statement approved. Each miss becomes a changed line in the plan with an owner and a date. | Drill clock | Inject | Expected answer | Pass condition | | --- | --- | --- | --- | | T+0:00 | The advisory arrives. CISA lists the CVE the same day | Event declared: the vendor states exploitation and the catalog lists it | Declared and logged within 1 hour, trigger named | | T+0:30 | "Which nodes do we run, and which patch level?" | Per node: 3.2 needs Patch 11, 3.4 needs Patch 7, 3.0 has no fix and needs a move to a supported release | Count from the real inventory within 2 hours | | T+1:00 | The network owner objects to access control lists on the ISE nodes | Pre-approved mitigation: allow only required management and control plane traffic. Cost: any admin station or integration not on the list loses access, so list the callers first | Mitigation applied within 4 hours, with its cost and an expiry in the log | | T+2:00 | The change approver does not answer the phone | Deputy approver, then the executive on call | Someone with authority is reached within 60 minutes | | T+3:00 | An application owner refuses a restart in business hours | Ring schedule or a signed acceptance with an expiry date | Decision logged with owner and expiry | | T+4:00 | On one node, the command show logging application ise-kong/access.log \| include dummyuser returns an entry for a suspicious username | Move to the Compromise flag, hand over to the incident response plan, check every node, cross-check firewall logs outside the device because Cisco says evidence may be removed, and re-image as Cisco recommends | Flag changed and evidence preserved before any re-image | | T+5:00 | A customer asks whether you are affected | Communications lead with legal, using confirmed facts only | Statement approved by the named person | ## An example drill result, constructed The result below is invented to show the form. It uses the invented card above (2 nodes on 3.2 Patch 9, 3 on 3.4 Patch 6 and 1 on 3.0) and is not a record of a real team. | Time compared | Clause 6 target | Constructed result | Plan line changed | Owner and date | | --- | --- | --- | --- | --- | | Declaration | 1 hour | 1 h 20 min: the analyst waited for the event lead, who was in a meeting | Clause 2: the deputy event lead may declare an Event on the catalog trigger alone | Security lead, 14 days | | First inventory answer | 2 hours | 3 h 10 min: patch level is not an inventory field, so each node was logged into by hand | Clause 5: add patch level to the inventory fields the first hour needs | Platform lead, 30 days | | First control | 4 hours | 3 h 40 min: met; the access-list pre-approval was used | None | None | | Approver reached | 60 minutes | 1 h 50 min: the deputy approver's number was out of date | Clause 3: the roster is call-tested each quarter | Security lead, 14 days | | Statement approved | Named person approves | 5 h 25 min: met, with the confirmed facts only | None | None | ## Related pages - [What to do when a zero day is disclosed: the first four hours, worked on PHP2Shell](/guides/what-to-do-when-a-zero-day-is-disclosed/) - [Compromise assessment after patching a critical flaw](/guides/compromise-assessment-after-patching/) - [Out-of-band patch: how to run one without breaking production](/guides/out-of-band-patch/) ## Sources - NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning: https://csrc.nist.gov/pubs/sp/800/40/r4/final (read 19 Nov 2026) - NIST SP 800-61 Rev. 3, Incident Response Recommendations and Considerations: https://csrc.nist.gov/pubs/sp/800/61/r3/final (read 19 Nov 2026) - Cisco advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - SOCRadar: SNOWLIGHT campaign, Tomcat exploitation dates: https://socradar.io/blog/snowlight-government-chinese-campaign/ (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/zero-day-response-plan/ --- # The first four hours after a zero day disclosure run in a fixed order, shown here hour by hour on PHP2Shell Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** In the first four hours after a zero day disclosure, extract seven facts from the advisory, check exploitation signals beyond the CISA catalog, find the web-facing binary on every host, and apply the vendor's mitigation order. The PHP2Shell release of 21 Oct 2026 is worked from the 14:00 UTC release to hour four. ## PHP2Shell is the running example: attack tools were in use about ten hours before the fix The PHP release of 21 Oct 2026 (CVE-2026-109618 and CVE-2026-109619, nicknamed PHP2Shell) allows unauthenticated remote code execution in PHP that serves web requests. Fastly's telemetry, published on 23 Oct 2026, shows a named tool at 03:59 UTC on 21 Oct, about ten hours before the coordinated release at 14:00 UTC, and CISA listed both bugs on 26 Oct. None of that was visible during the four hours below. The steps for the whole 72 hours on this asset type are in the [application runtime playbook](/playbooks/application-runtime/). A team that read the 19:30 pre-announcement on 20 Oct had 18 hours 30 minutes of warning. A team that first read the advisory at 14:00 started about ten hours after the first named tool. ## Hour 0 to 1: write seven facts from the advisory into the record The security lead opens the vendor advisory, GHSA-w8hw-w9fx-j79w, and fills seven fields. The record leaves a one-line answer per field, each with the time it was read. | Field | PHP2Shell, from the advisory | | --- | --- | | Disclosure time | Release 21 Oct 2026 at 14:00 UTC; GitHub advisory published 21 Oct | | Affected | PHP 8.2 before 8.2.35, 8.3 before 8.3.36, 8.4 before 8.4.27, 8.5 before 8.5.12 | | No fix exists for | PHP 8.1 and older (no upstream fix) | | Exposed through | Apache mod_php (rated highest by PHP), PHP-FPM, php-cgi, FrankenPHP, LiteSpeed | | Workaround | None that is complete: updating is the only complete mitigation. memory_limit=-1 with a restart of every worker is partial | | Exploitation statement | The advisory confirms remote code execution over HTTP against mod_php and says nothing about attacks in the wild | | Companion flaw | CVE-2026-109619 comes from the same release; the advisory says to follow the remediation advice for both | ## Hour 1 to 2: an empty catalog is not a clear signal, so read two outside sources and your own logs A security lead who searched the CISA catalog at 15:00 on 21 Oct found nothing, because the listing came on 26 Oct. The advisory has no attack statement, so the signals come from elsewhere. Record the result of each with a time. Fastly published the request shape on 23 Oct, so on 21 Oct a team had only the first source below and its own logs. Search back to the pre-announcement of 20 Oct, 19:30 UTC. - CISA's alert of 21 Oct, Critical Vulnerabilities in PHP Require Immediate Patching. It names the fixed releases, tells organizations to inventory every installation, including bundled copies, and says to monitor for PHP worker processes launching shells or downloading tools. CISA added a SIGMA detection rule on 23 Oct. - Fastly's report of 23 Oct, which names the request shape of the most widespread exploit kit: POST to / or /index.php, Content-Type multipart/form-data with the fixed boundary B7kQpoc618, a fixed-size body, a minimal header set, raw IP addresses as targets, one or two requests per host. Fastly says the boundary string is the most reliable separator; User-Agent strings (php2shell-109618 here), source IPs and body sizes go stale within a day. - Your own edge logs, searched for that shape. Success is not visible at the edge: success shows at the origin, in PHP-FPM and Apache error logs (worker crashes such as AH00051 child exit entries) and in outbound connections from PHP workers. A hit changes the plan: start the compromise check on that host before you patch or rebuild it, and keep the evidence first. - Look-alikes, ruled out before anything is counted. Fastly measured 98.6% and 99.9% of two circulating PHP2Shell signatures (chunked multipart with no Content-Length, bodies padded with the letter A) as React2Shell traffic, which carries a Next-Action header that PHP traffic never does. A WordPress kit whose boundary starts ----wp2shell is a separate July 2026 campaign. ## Hour 2 to 3: find the web binary, not the command-line tool Platform engineering answers where the vulnerable component serves requests. A command-line tool on a host can report a different version from the one behind the web server, so the check runs against the running service. For PHP that means the web server API (SAPI). Group the answer by what the host does, with the server type the vendor rates highest first (PHP rates Apache mod_php highest). A version with no upstream fix is a group of its own: the answer is a distributor backport or an upgrade. - Running web servers: ps -ef | grep -Ei 'php-fpm|php-cgi|frankenphp|lsphp|httpd|apache2' | grep -v grep. Whether Apache loads mod_php: apachectl -M | grep -i php (apache2ctl -M on Debian and Ubuntu). Packages: dpkg -l 'php*' 'libapache2-mod-php*' or rpm -qa 'php*'. A host that runs PHP only from the command line, for cron jobs, drops out of web scope. - Version of the web binary: php-fpm -v or php-cgi -v; for mod_php, php -v after confirming the module. Distributions backport fixes without changing the number, so ask the package: rpm -q --changelog php | grep -i 109618 or apt changelog php8.3 | grep -i 109618, and the same with 109619. A version below the fixed release can still be patched. Fixed upstream releases are 8.2.35, 8.3.36, 8.4.27 and 8.5.12; 8.1 and older have none. - Check every container image and every product that bundles the component. - Leave one line per host or image: server type, version, changelog hit yes or no, reachable from the internet yes or no, owner. ## Hour 3 to 4: apply the vendor's mitigation order and write each cost down Patching needs a canary first. While it runs, apply the options in order of how much they remove, and write each into the record with its cost, its owner and the time it will be reviewed. The options below are PHP's and CISA's own, except the frontend refusal, which is this site's rule for applications with no uploads. | Option | What to set | Stated cost or limit | Source | | --- | --- | --- | --- | | 1. Update | PHP 8.2.35, 8.3.36, 8.4.27 or 8.5.12, or the distribution's patched package; restart every worker | A staging test first; the fix changes out-of-memory handling. A package install does not replace running workers | PHP advisory GHSA-w8hw-w9fx-j79w: the only complete mitigation | | 2. Partial setting change | CVE-2026-109618: memory_limit=-1 as an admin value (php_admin_value[memory_limit] = -1 in an FPM pool; php_admin_value memory_limit -1 in Apache), then restart every worker. CVE-2026-109619: session.upload_progress.enabled=0, and on PHP 8.5 fatal_error_backtraces=0 | A runaway script can use all host memory; upload progress stops. Partial: a failed mmap() can still reach the vulnerable path. Do not use RLIMIT_AS or RLIMIT_DATA, which make mmap() fail | PHP advisories GHSA-w8hw-w9fx-j79w and GHSA-6wmf-4rv5-p3rp | | 3. Refuse at the frontend | Keep post_max_size conservative and enforce the decoded size at the frontend. Where the application has no uploads, refuse multipart before PHP, for example in nginx: if ($content_type ~* "^multipart/form-data") { return 415; } | File uploads and multipart forms stop working | PHP advisory (size limits); the 415 rule is this site's | | 4. WAF rule | A rule for both CVEs, as a stopgap; Fastly Next-Gen WAF has one | Partial. Fastly warns that circulating IP block lists include VPN exits, Tor exits and residential addresses carrying ordinary traffic | CISA alert; Fastly report | | 5. Take it offline | Systems that cannot be updated, and PHP 8.1 and earlier | The service is down; accept in writing with an expiry date if it must stay up | CISA alert | ## Hour 4: log one decision per group and set the next update time Close the first window with a decision per group, an owner and a review time. The entries below show the form; the group names are examples to replace with your own. The log feeds the board brief and the 72-hour review. | Group (example) | Decision at 18:00 UTC | Owner | Review | | --- | --- | --- | --- | | Public web tier on Apache mod_php | Patch in rings starting tonight; memory_limit=-1 until the fixed build is verified | Platform engineering | 22:00 UTC, next update | | Internal tools on PHP-FPM | Patch within 24 hours | Application owner | 22 Oct 14:00 UTC | | Hosts on PHP 8.1 | Take offline, or accept in writing with an expiry date | Executive | Expiry date required | ## Five situations where the first four hours change The PHP order assumes a fix, a workaround and no earlier compromise. These cases from the site's CVE data break one of those assumptions. - Exploitation began before the advisory. BeyondTrust's CVE-2026-102764 was used from 29 Aug 2026 and disclosed on 15 Oct. Start with the compromise check: the actor deleted local logs, so off-box syslog is often the only record, and BeyondTrust and Mandiant advise rebuilding from a clean image rather than cleaning in place. - No workaround exists. Palo Alto lists none for PAN-OS CVE-2026-0322, and Cisco lists none for ISE CVE-2026-76460. Hour 3 becomes access restriction: a jump box for the PAN-OS management interface, which Palo Alto rates 8.5 instead of 8.6, so it lowers risk only a little, and infrastructure access control lists for ISE. - The vendor withholds details. BeyondTrust held back technical details while self-hosted appliances were patched, and Mandiant notes the actor rotates infrastructure per victim. Hunt for behavior (unrecognized privileged sessions, scripts in web directories), not for one victim's indicators. - Some copies are patched for you. BeyondTrust patched its SaaS instances automatically on 8 Oct 2026; self-hosted appliances needed action. Ask the vendor which copies it handled. - A version has no fix. Cisco ISE 3.0 is past software maintenance and has no patch in the 16 Sep 2026 advisory; the answer is a move to a supported release. ## Related pages - [Out-of-band patch: how to run one without breaking production](/guides/out-of-band-patch/) - [Zero day response plan: the complete template with owners, triggers and a drill](/guides/zero-day-response-plan/) - [The board brief by noon: template, interim note and a filled brief for Cisco ISE](/guides/board-brief-by-noon/) ## Sources - PHP security advisory GHSA-w8hw-w9fx-j79w: https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 19 Nov 2026) - Fastly: PHP2Shell timeline, telemetry and indicators (23 Oct 2026): https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell (read 19 Nov 2026) - CISA alert: Critical Vulnerabilities in PHP Require Immediate Patching: https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching (read 19 Nov 2026) - BeyondTrust: ARMORY CASTLE and BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - Mandiant: ARMORY CASTLE Exploits BeyondTrust: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - Palo Alto Networks advisory CVE-2026-0322: https://security.paloaltonetworks.com/CVE-2026-0322 (read 19 Nov 2026) - Cisco advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/what-to-do-when-a-zero-day-is-disclosed/ --- # Run an out-of-band patch in rings, with a rollback that keeps the mitigation in place Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** An out-of-band patch is a fix applied outside the maintenance calendar because exploitation will not wait. Run it with a named approver, a smoke test the application owner wrote, small rings and a rollback that keeps any mitigation. JFrog's Artifactory flaw went from advisory on 28 Aug 2026 to about 406,000 attempts a day by 2 Sep. ## Go out of band when the calendar is slower than the attacker NIST SP 800-40 Rev. 4 defines enterprise patch management as "the process of identifying, prioritizing, acquiring, installing, and verifying the installation of patches, updates, and upgrades throughout an organization." An out-of-band patch runs that process outside the calendar, so every part of it has to exist before the day it is needed. CISA's BOD 26-04 builds its federal deadlines from four inputs: whether the asset is publicly exposed, whether the CVE is in the KEV catalog, whether an attacker can automate the exploit, and the technical impact. The directive says that of the two events that can start a timeline, "Whichever event occurs first starts the remediation timeline." Deadlines in the KEV entries this site holds are not uniform, so read the due date of each one. Some flaws arrive already exploited. Microsoft marked the Windows AFD flaw CVE-2026-68820 as exploited on 11 Aug 2026, the day it published the fix, and CISA listed it the same day. PHP2Shell was earlier still: Fastly saw the first request from a PHP2Shell-named tool at 03:59 UTC on 21 Oct 2026, 10 hours 1 minute before PHP's coordinated release at 14:00 UTC. | CVE | Added to KEV | KEV due date | Days allowed | | --- | --- | --- | --- | | CVE-2026-82329, Artifactory | 2 Sep 2026 | 5 Sep 2026 | 3 | | CVE-2026-0322, PAN-OS | 14 Oct 2026 | 17 Oct 2026 | 3 | | CVE-2026-76418, Cisco ASA and FTD | 29 Oct 2026 | 1 Nov 2026 | 3 | | CVE-2026-123158, OpenSSH | 9 Nov 2026 | 12 Nov 2026 | 3 | | CVE-2026-111497, Node.js and Brotli | 27 Oct 2026 | 10 Nov 2026 | 14 | | CVE-2026-41709, VMware ESX | 28 Oct 2026 | 11 Nov 2026 | 14 | | CVE-2026-19868, SonicWall SonicOS | 6 Nov 2026 | 20 Nov 2026 | 14 | ## Name the approver, the deputy and the owner of each gate before the day The change record needs four entries: the advisory link as read, the scope, the rollback and the owner of the smoke test. The approver has emergency change authority, so the change board is not a gate; name the approver, a deputy and the channel in advance. The application owner holds what platform engineering lacks: which login, checkout or API call proves the service works. ## Write the smoke test and the rollback first, then widen in rings A smoke test runs in under ten minutes and ends in a pass or a fail that anyone on call can read. A rollback is a snapshot of the host or volume together with its database, a previous image tag, or a pinned previous package. Take it before ring 0, and try it once on ring 0 hosts. A rollback of a security fix puts the flaw back. If the vendor gave a mitigation, apply it to the hosts first and keep it through a rollback, so the rolled-back host is still closed. Where no mitigation exists, a rollback is a decision for the approver, taken with the exposure in front of them. Keep rings small enough to roll back. The first ring is one or two hosts that carry real traffic and have owners who answer the phone. | Ring | Scope | Hold | Pass condition | | --- | --- | --- | --- | | 0 | Staging, or two hosts that carry real traffic | 30 minutes | Service starts, smoke test passes, version shows the fixed build | | 1 | About ten percent of exposed hosts | 1 hour | Error rate and latency match the baseline of the previous hour | | 2 | All other exposed hosts | Until each host is verified | Smoke test passes, fixed build recorded per host | | 3 | Internal hosts | Same day | Fixed build recorded, rescan clean | ## Plan for what must restart, and for versions with no fix An upgrade that leaves the old process running has not fixed anything. Check the start time of the running process against the install time: on Linux, ps -o lstart= -p $(pgrep -o sshd) prints when the oldest sshd started. Some versions get no patch. PHP 8.1 and older receive no upstream fix, and the Node.js advisory covers only the 22, 24 and 26 lines, so 20.x, 23.x and 25.x need a move to a supported line. Fortinet tells FortiMail 7.0 users to migrate to a fixed release. For such hosts the answer is the vendor mitigation or taking the host offline, which is what CISA's alert advises for PHP versions that cannot be updated. | Product and CVE | What restarts | Source note | | --- | --- | --- | | OpenSSH on Alpine, CVE-2026-123158 | rc-service sshd restart; existing sessions keep running | Alpine: upgrading alone does not replace the running daemon | | PHP, CVE-2026-109618 | Every PHP worker | PHP: a restart and regression testing for a minor release | | Node.js, CVE-2026-111497 | Rebuild images, then restart processes | Node.js advisory; copies of Brotli outside Node.js need their own upgrade | | Apache Tomcat cluster, CVE-2026-34486 | Each node, as a rolling restart if sessions replicate | Tomcat 11.0.21, 10.1.54 or 9.0.117 or later | | Linux kernel, CVE-2026-46300 | Reboot, or kpatch where Red Hat ships it | Red Hat offers kpatch for some streams | | VMware ESX, CVE-2026-41709 | Maintenance mode and a reboot per host | Broadcom: plan vMotion or a rolling cluster update | | FortiMail, CVE-2026-104286 | Reboot, with a mail-flow interruption | Fortinet: back up the configuration first | | ScreenConnect, CVE-2026-84869 | Host clients are reinstalled and access agents updated | ConnectWise: the server upgrade does not update technician clients | ## Worked case: Artifactory from Fri 28 Aug to Mon 31 Aug 2026 CVE-2026-82329 let an unauthenticated caller obtain an administrator token from the Artifactory Access service on a default install, where the join key is empty. The CNA's CVSS base score is 9.8. JFrog published six fixed builds on 28 Aug 2026, a Friday: 7.111.21, 7.117.28, 7.125.20, 7.133.29, 7.146.38 and 7.161.20. Fastly saw no exploitation on 28 Aug and only single-digit probes on 29 and 30 Aug. On Monday 31 Aug it recorded about 75,000 requests, nearly 98 percent from offensive-security services. A public exploit appeared on Tuesday 1 Sep, with about 171,000 attempts. On Wednesday 2 Sep, when CISA listed the CVE, attempts peaked near 406,000. The KEV due date was Saturday 5 Sep. The schedule below is built from those dates. Patching ends on Sun 30 Aug, before the Monday wave, and Monday is verification. The KEV due date of Sat 5 Sep is the outer limit, not the plan. - Fri 28 Aug, hour 0: list every self-hosted instance and read GET /artifactory/api/system/version on each. Match each to its train and fixed build. 7.146.37 is not enough; JFrog's affected range ends at 7.146.36 but NVD treats everything below 7.146.38 as affected. Look for additionalJoinKeys under shared, security in system.yaml. - Hour 2: on instances reachable from the internet, apply JFrog's workaround. Generate a key with openssl rand -hex 16, add it as additionalJoinKeys, keep the existing joinKey, and restart the Access service. The existing key keeps working, so cluster peers are unaffected. The mitigation also stays in place if a later upgrade is rolled back. - Hour 4: ring 0, the staging instance, to the first fixed build for its train. Smoke test: the version endpoint shows the build, one CI job deploys and resolves a small artifact with a real service token, and a second cluster node rejoins with the existing join key. - Sat 29 Aug: ring 1, one production instance serving the team with the lowest release pressure, held for an hour. Sun 30 Aug: ring 2, all other instances, one per train at a time. - Mon 31 Aug: verify the version on every instance. Then run the hunt in the compromise assessment guide (Compromise assessment after patching) for the window from 28 Aug. - Same day: patching does not revoke tokens already issued. Rotate the platform join key and revoke access tokens issued since 28 Aug. ## Close the change with someone who did not deploy it Close an out-of-band change only when a second person, or the scanner, has read the fixed version on each asset. A deployment log shows that a command ran, not what is running. List assets still open with an owner and a date, and remove each temporary mitigation after the check, with the removal logged. Patching changes the code, not what an attacker already took. ConnectWise told all partners on 13 Nov 2026 to verify that every technician client runs 26.6.5 or later, because updating the console does not update clients. Where the CVE is in the KEV catalog with forensic triage marked Yes, as FortiMail's is, preserve the appliance logs and a configuration backup before changing the appliance. Keep for the auditor: the change number, approver, the advisory text as read with its URL and read date, the ring plan, the smoke test result per ring, the fixed build per asset, the verification output, and a time for each step. The same record feeds the 72-hour review. ## Related pages - [The board brief by noon: template, interim note and a filled brief for Cisco ISE](/guides/board-brief-by-noon/) - [What to do when a zero day is disclosed: the first four hours, worked on PHP2Shell](/guides/what-to-do-when-a-zero-day-is-disclosed/) - [Zero day vulnerability defined: the sources, the tests and six 2026 cases](/guides/zero-day-vulnerability-defined/) ## Sources - NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning: https://csrc.nist.gov/pubs/sp/800/40/r4/final (read 19 Nov 2026) - CISA BOD 26-04, Prioritizing Security Updates Based on Risk: https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk (read 19 Nov 2026) - JFrog security advisories, CVE-2026-82329 section and workaround: https://docs.jfrog.com/releases/docs/jfrog-security-advisories (read 19 Nov 2026) - Fastly Security Research, Artifactory exploitation activity, 3 Sep 2026: https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity (read 19 Nov 2026) - JFrog documentation, Get Artifactory Version API: https://docs.jfrog.com/administration/reference/getartifactoryversion (read 19 Nov 2026) - Microsoft MSRC CVRF, August 2026 release: https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug (read 19 Nov 2026) - Microsoft MSRC record for CVE-2026-68820: https://api.msrc.microsoft.com/sug/v2.0/en-US/vulnerability/CVE-2026-68820 (read 19 Nov 2026) - Fastly report on PHP2Shell exploitation and disclosure timeline: https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell (read 19 Nov 2026) - CISA alert, Critical Vulnerabilities in PHP Require Immediate Patching: https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching (read 19 Nov 2026) - PHP security advisory GHSA-w8hw-w9fx-j79w: https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 19 Nov 2026) - Alpine Linux advisory, OpenSSH password authentication bypass, 5 Nov 2026: https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/ (read 19 Nov 2026) - Node.js security releases, 21 Oct 2026: https://nodejs.org/en/blog/vulnerability/october-2026-security-releases (read 19 Nov 2026) - Apache Tomcat 11 security page: https://tomcat.apache.org/security-11.html (read 19 Nov 2026) - Red Hat CVE page for CVE-2026-46300: https://access.redhat.com/security/cve/CVE-2026-46300 (read 19 Nov 2026) - Broadcom security advisory VMSA-2026-0006: https://support.broadcom.com/web/ecx/support-content-notification/-/external/content/SecurityAdvisories/0/38017 (read 19 Nov 2026) - Fortinet PSIRT advisory FG-IR-26-175: https://fortiguard.fortinet.com/psirt/FG-IR-26-175 (read 19 Nov 2026) - ConnectWise ScreenConnect security bulletin, 8 Sep 2026: https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin (read 19 Nov 2026) - ConnectWise advisories page: https://www.connectwise.com/company/trust/advisories (read 19 Nov 2026) - CISA KEV entry for CVE-2026-82329: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-82329 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-0322: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-76418: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76418 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-123158: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-123158 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-111497: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-111497 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-41709: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-41709 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-19868: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19868 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-68820: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-68820 (read 19 Nov 2026) - NVD record for CVE-2026-82329: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-82329 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/out-of-band-patch/ --- # The board brief by noon has five lines, a stated next update time and a rule for when the advisory lands late Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** Five lines fill the board brief: what is affected, whether exploitation is confirmed, what is done, what is open, and when the next update comes. Below are the brief and interim note templates, a brief filled from the Cisco ISE advisory of 16 Sep 2026 at 16:00 GMT, and the notification clocks. ## The brief template Directors have two minutes. Write one sentence per line, in this order. Copy the form and fill every field; write "unknown, answer by HH:MM" where a field is not yet known. The brief is due at 12:00 in the time zone the board meets in, written in the subject line; every time in the brief uses that zone. Which day depends on when the advisory is public. If it is public before 10:00 that day, the full brief is due at noon the same day. From 10:00 on, fewer than two hours remain, so the advisory counts as late: send the interim note within two hours of the advisory and the full brief at noon the next day. - Subject: [Product] flaw [CVE], brief [n], [date and time, time zone]. - 1. Affected: [product and versions] on [number] systems, of which [number] are reachable from the internet or partner networks. - 2. Exploitation: [confirmed on our systems, confirmed in the wild, or not known], per [named source], dated [date]. Compromise assessment: [open until HH:MM, or closed with the finding]. - 3. Done: [controls applied] on [number] of [number] systems at [HH:MM]; patched [number] of [number]. - 4. Open: [number] systems not yet fixed, owner [role], fix by [date and time]. - 5. Next update: [HH:MM, time zone]. - Decisions needed: [authority for an outage window, a risk acceptance with an expiry date, spend, or none]. ## The interim note template When the advisory is late, as defined above, the interim note keeps the same five lines with unknowns marked and goes out within two hours of the advisory. Use it for the hours before the first full brief: - Subject: [Product] flaw [CVE], interim note, [date and time]. - 1. Affected: scoping under way; asset list due [HH:MM]. - 2. Exploitation: [vendor or CISA statement and date, or not known]. - 3. Done: [first control, or none yet; first control due HH:MM]. - 4. Open: everything in scope until the list exists. - 5. Next update: the full brief by noon on [date]. ## A filled brief: Cisco ISE, CVE-2026-76460 Cisco published advisory cisco-sa-ISE-ABP-VNSW7Tn5 on Wednesday 16 Sep 2026 at 16:00 GMT. The facts that fill the brief come from the advisory, NVD and CISA. Fields marked in square brackets are counts only your inventory can supply, and no public record can fill them. Because 16:00 GMT is after the 10:00 cut-off, the rule above applies to a board that meets in GMT: send the interim note by 18:00 GMT on 16 Sep and the full brief by noon on 17 Sep. The interim note comes first below, then the noon brief. - Interim note, 16 Sep 2026: - Subject: Cisco ISE flaw CVE-2026-76460, interim note, 16 Sep 2026, 18:00 GMT. - 1. Affected: scoping under way; Cisco says ISE and ISE-PIC are affected regardless of configuration; node list due 20:00 GMT. - 2. Exploitation: Cisco states its PSIRT is aware of active exploitation; CISA added the bug to its catalog on 16 Sep 2026. - 3. Done: nothing yet; Cisco lists no workaround, and the first control, infrastructure access lists on the nodes, is due 22:00 GMT. - 4. Open: every ISE node until the list exists. - 5. Next update: the full brief by noon on 17 Sep 2026. - Noon brief, 17 Sep 2026: - Subject: Cisco ISE flaw CVE-2026-76460, brief 1, 17 Sep 2026, 12:00 GMT. - 1. Affected: Cisco ISE and ISE-PIC on every release below 3.1 Patch 12, 3.2 Patch 11, 3.3 Patch 12, 3.4 Patch 7 or 3.5 Patch 4, regardless of configuration, on [number] nodes, [number] reachable from partner networks; and ISE 3.0, which is end of software maintenance and has no fix, on [number] nodes. - 2. Exploitation: confirmed in the wild. Cisco states its PSIRT is aware of active exploitation and gives no start date. CISA added the bug to its catalog on 16 Sep 2026. Compromise assessment of the [number] nodes: open until [HH:MM]; Cisco's check is the ise-kong access log on every node. - 3. Done: Cisco has no workaround, so the control is infrastructure access lists that allow only required management and control plane traffic; applied on [number] of [number] nodes at [HH:MM]. Patched: [number] of [number]. - 4. Open: [number] nodes not yet patched, owner [role], fix by [date and time]. CISA's due date for federal agencies is 19 Sep 2026, three days after the advisory. - 5. Next update: [HH:MM] GMT on 17 Sep 2026. - Decisions needed: authority to re-image any node the assessment flags, which Cisco strongly recommends if compromise is suspected; approval of the access-list change that cuts off any admin station or integration not on the list; and a migration date for each ISE 3.0 node, because Cisco advises moving to a supported release that includes the fix, with the access lists as the stopgap until then. ## Say what is unknown, and ask for decisions Directors accept uncertainty when it is labeled and dated. They do not accept a surprise. Do not write that there is no impact until the assessment has closed; write "no evidence found as of HH:MM in [the sources checked]; assessment open". BeyondTrust's CVE-2026-102764 shows why: exploitation began on 29 Aug 2026 and the advisory came on 15 Oct 2026, so a brief based on the advisory alone would have missed 47 days. End with the decisions you need, or the word none. A brief that asks for nothing and names the next update time lets the board stop thinking about it until then. Keep CVSS out unless a director asks; exposure and exploitation matter more to the decision. ## Who supplies each line, who approves it, and how the brief fails The security lead drafts, using the scribe's decision log. The executive approver reads and releases it; the communications lead checks line 4 against the notification duties with legal. The sent brief is filed in the event's decision log with its send time, so brief 2 can be compared with brief 1. | Line | Supplied by | Source of the number | How it typically fails, and the fix | | --- | --- | --- | --- | | 1. Affected | Platform lead | Inventory export, with its time | A stale count: the inventory was last refreshed days ago. Print "as of HH:MM" beside every count and recount before sending | | 2. Exploitation | Event lead | Vendor advisory and CISA catalog, with the date of each; assessment result from the SOC lead | Premature "no impact". Keep the wording above until the assessment closes | | 3. Done | Platform lead | Verification output from the hosts, not the change ticket | The control is reported as applied when it was only requested. Count only hosts where the check in the mitigation guide passed | | 4. Open | Named owner of each open system | Decision log entry with the fix-by time | A missing owner: the line names a team, so nobody answers the director. Name a person, or do not send | | 5. Next update | Event lead | The update interval in the plan | The time is missed and nobody says so. If the update cannot be ready, send a one-line note before the stated time with a new time | | Decisions needed | Executive approver, asked by the event lead | The authority table in the response plan | The question is buried in line 3. Keep it as its own line, with the expiry of any acceptance | ## Brief 2 when compromise is found on your own systems The five lines stay, but the meaning of three of them changes, and the hand-over to the incident response plan is itself an item in the brief. - Subject: [Product] flaw [CVE], brief [n], compromise confirmed, [date and time, time zone]. - 1. Affected: [number] systems with evidence of compromise, first evidence at [time] in [named log]; [number] more in scope and not yet assessed. - 2. Exploitation: confirmed on our systems. Evidence preserved before any patch or re-image: yes or no, by [role]. The incident response plan took over at [HH:MM], incident lead [name]. - 3. Done: [number] systems isolated, [number] credentials and tokens reset, [number] systems rebuilt; mitigation and patch status as in brief 1. - 4. Open: assessment of the remaining [number] systems, owner [role], done by [date and time]. Notification duties: awareness recorded at [HH:MM]; each applicable clock and its deadline, from the table below, with the owner of the decision to notify. - 5. Next update: [HH:MM, time zone], at shorter intervals than before. - Decisions needed: authority to rebuild, to notify regulators or customers, and to take the service offline. ## Notification clocks that may start the day the brief is written These apply only if their trigger is met. Put the one that applies, with its deadline, in line 4 of the brief. | Rule | Trigger | Deadline | Source | | --- | --- | --- | --- | | GDPR Article 33 | A personal data breach, unless unlikely to risk people's rights and freedoms | "without undue delay and, where feasible, not later than 72 hours after having become aware of it"; delay needs reasons | gdpr-info.eu, Art. 33 | | NIS2 Directive (EU) 2022/2555 | A significant incident at an essential or important entity | Early warning without undue delay and within 24 hours; incident notification within 72 hours of becoming aware | EUR-Lex, Directive (EU) 2022/2555 | | SEC Form 8-K Item 1.05 | A US-listed registrant determines that a cybersecurity incident is material | "generally" due "four business days after a registrant determines" that; the clock runs from the determination, not from discovery | SEC press release 2023-139 | ## The record behind the brief If a director or auditor asks how you knew, the decision log answers: the advisory link and read time, each decision with its approver and time, the verification result, and each brief as sent. NIST SP 800-61r3 calls incident response "a critical part of cybersecurity risk management that should be integrated across organizational operations", which is where the brief lands. Have an executive read the draft aloud once before it goes out. A brief that reads badly aloud is usually missing a number or an owner. ## Related pages - [Zero day vulnerability defined: the sources, the tests and six 2026 cases](/guides/zero-day-vulnerability-defined/) - [Out-of-band patch: how to run one without breaking production](/guides/out-of-band-patch/) - [Mitigations that buy time before the patch](/guides/mitigations-that-buy-time/) ## Sources - Cisco Security Advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog entry for CVE-2026-76460: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460 (read 19 Nov 2026) - NVD record for CVE-2026-76460: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76460 (read 19 Nov 2026) - BeyondTrust blog on BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - GDPR Article 33: https://gdpr-info.eu/art-33-gdpr/ (read 19 Nov 2026) - Directive (EU) 2022/2555 (NIS2): https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32022L2555 (read 19 Nov 2026) - NIST SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cyber Risk Management: https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-61r3.pdf (read 19 Nov 2026) - SEC press release 2023-139, cybersecurity risk management and incident disclosure rules: https://www.sec.gov/newsroom/press-releases/2023-139 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/board-brief-by-noon/ --- # A zero day vulnerability is a flaw exploited before its fix is public, and the label depends on which dates you can see Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** Google's threat intelligence group defines a zero-day as a vulnerability "maliciously exploited in the wild before a patch was made publicly available." NIST's glossary says "previously unknown" instead. The two tests can disagree on one bug, so this guide applies them to six 2026 CVEs from the site's data; the [glossary](/glossary/) defines other terms. ## Four sources use the term four ways NIST's entry is about an attack, the others about a vulnerability. NIST's words do not say unknown to whom: the vendor, the public or the defender. A defender cannot see the vendor's knowledge, which is why the working tests below rest on dates. | Source | Wording | Test it implies | | --- | --- | --- | | NIST CSRC glossary, "zero-day attack" (from CNSSI 4009-2022 and NISTIR 8011 Vol. 3) | "An attack that exploits a previously unknown hardware, firmware, or software vulnerability." | Test A: the flaw was unknown when it was first used | | Google Threat Intelligence Group, 2024 zero-day analysis | "GTIG defines a zero-day as a vulnerability that was maliciously exploited in the wild before a patch was made publicly available." | Test B: first exploitation is earlier than the public fix | | Microsoft Security Update Guide record | Two fields at release, publiclyDisclosed and exploited, and a release status such as "Exploitation Detected" | Test C: the vendor marks exploitation on the day the fix ships | | Project Zero, vendor deadlines | Not a definition. For evidence of exploitation against real users "in the wild", details are disclosed after 7 days, against a normal 90 | Evidence of use in the wild shortens the vendor's time to fix | ## Three dates decide which test a bug passes Collect three dates from the advisory and the reports around it: E, the earliest dated evidence of exploitation; V, when the vendor learned of the flaw; F, when a fix became public. - E earlier than F: a zero day under Test B. - E earlier than V: the strict sense of Test A, because the vendor could not have known. V is rarely public, so it usually comes from the vendor's own account. - E after F: not a zero day under either test. It is an n-day, a flaw exploited after its fix is public. - E on the release day with no earlier date: record it as exploited at release, earlier use not shown, and do not call it either way. ## BeyondTrust CVE-2026-102764 shows E, V and F for one bug BeyondTrust's own timeline gives every date. Counted from E, 29 Aug 2026: the vendor first saw evidence after 30 days, the patch came after 40 days and disclosure after 47 days. The same bug is a zero day under Test A for 30 days and under Test B for 40. A third reading applies to a self-hosted customer. The patch existed from 8 Oct, but a customer who had not been told had no reason to apply it until 15 Oct. The zero day ends for you on the day you learn there is something to apply. | Date (2026) | Event | Role | | --- | --- | --- | | 29 Aug | The actor exploits the flaw against one BeyondTrust-hosted instance and one self-hosted appliance | E | | 27 Sep | Another vendor's zero day becomes public, which starts a hunt in BeyondTrust's threat intelligence platform | | | 28 Sep | The hunt finds the hosted instance; it is contained within 27 minutes of detection; Product Security reproduces the entry route that evening | V | | 30 Sep | Mandiant shares findings from the self-hosted customer; BeyondTrust validates a fix on a test build | | | 8 Oct | BT26-05 patch released to SaaS and self-hosted customers; SaaS instances patched automatically | F | | 15 Oct | Coordinated disclosure of CVE-2026-102764 and the public advisory | Disclosure | ## Adobe's Style Smuggler was exploited three days before its hotfix Sansec reports that attacks on CVE-2026-75650 started on 4 Sep 2026. It published its findings on 5 Sep and wrote that the attackers were iterating quickly. Adobe published APSB26-146 on 7 Sep at 20:20 UTC, with the fix shipped as hotfix VULN-39341 rather than a new release number. E is 4 Sep and F is 7 Sep, so the bug passes Test B with three days of margin. Unlike the BeyondTrust case, an outside firm watching stores supplied the evidence, and the public record gives no V. ## Four more cases, each read against the tests The record decides the label, and the label should say what the record shows. The PHP row passes Test B on attempts only; the other three do not pass it. | CVE and product | Fix or advisory | Earliest exploitation record | What the record supports | | --- | --- | --- | --- | | CVE-2026-109618, PHP | 21 Oct 2026, 14:00 UTC | 21 Oct, 03:59 UTC: first request from a named tool; 05:11 UTC: first exploit kit (Fastly telemetry) | Attack tools about ten hours before the fix. Fastly notes that success is not visible at the edge, so the evidence is of attempts | | CVE-2026-68820, Windows afd.sys | 11 Aug 2026 | 11 Aug: Microsoft marks exploited Yes and publicly disclosed No at release | Exploited at release; earlier use not shown, so Test B cannot be applied | | CVE-2026-34486, Apache Tomcat | 9 Apr 2026 | 24 Apr, the earliest date in SOCRadar's range of 24 to 29 Apr | An n-day, exploited 15 days after the advisory | | CVE-2026-9198, Langflow | Release 1.10.1 on 23 Jun 2026; IBM bulletin 2 Jul | CISA's entry in NVD, stamped 17 Jul, lists exploitation as active; catalog listing 4 Aug | An n-day: the fix was public weeks before any evidence of use | ## Where the usual answer is wrong - Unknown to the vendor is not something a runbook can test. Write the trigger in observable terms: a dated report of exploitation, a fix date, and the gap between them. - A low rating does not rule out exploitation, and a high one does not make a zero day. Broadcom rates ESX CVE-2026-41709 Low (CVSS 2.7) and CISA still added it to its catalog on 28 Oct 2026. The rating describes the flaw; the dates describe the event. - This site uses the label in observable terms only. The Disclosure Clock on the [home page](/) records disclosure, first dated exploitation and the days between, and does not call a bug a zero day; the playbooks start their hour count at the advisory, so for a bug exploited earlier, Question 6 of the [72-hour review](/guides/72-hour-post-incident-review/) sets the window to search. - A patch on the shelf does not end the exposure for hosts that have not applied it. BeyondTrust's SaaS customers were patched on 8 Oct without action; self-hosted appliances were not. ## Related pages - [Mitigations that buy time before the patch](/guides/mitigations-that-buy-time/) - [The board brief by noon: template, interim note and a filled brief for Cisco ISE](/guides/board-brief-by-noon/) - [How to tell where a new CVE hits you](/guides/where-a-new-cve-hits-you/) ## Sources - NIST CSRC glossary: zero-day attack: https://csrc.nist.gov/glossary/term/zero_day_attack (read 19 Nov 2026) - Google Threat Intelligence Group: A 2024 Zero-Day Exploitation Analysis: https://cloud.google.com/blog/topics/threat-intelligence/2024-zero-day-trends (read 19 Nov 2026) - Microsoft Security Update Guide record for CVE-2026-68820: https://api.msrc.microsoft.com/sug/v2.0/en-US/vulnerability/CVE-2026-68820 (read 19 Nov 2026) - Project Zero: deadlines for vendors: https://projectzero.google/vulnerability-disclosure-policy.html (read 19 Nov 2026) - BeyondTrust: ARMORY CASTLE and BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - Mandiant: ARMORY CASTLE Exploits BeyondTrust: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - Sansec: StyleSmuggler research: https://sansec.io/research/stylesmuggler-0day (read 19 Nov 2026) - Fastly: PHP2Shell timeline and telemetry: https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell (read 19 Nov 2026) - SOCRadar: SNOWLIGHT campaign: https://socradar.io/blog/snowlight-government-chinese-campaign/ (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - IBM security bulletin for Langflow OSS: https://www.ibm.com/support/pages/node/7278927 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/zero-day-vulnerability-defined/ --- # A mitigation buys hours, not a fix, so pick the cheapest one that closes the path and prove it took Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** Mitigations that buy time fall into a few types: restrict the interface, disable the feature, filter at a proxy or WAF, take the service off the internet, and change configuration. Each costs something and each can be checked. OpenSSH's fix went public on 5 Nov 2026 and the first exploitation came 39 hours 46 minutes later. ## Ten mitigations from 2026 advisories, what each costs and how to prove it took NIST SP 800-40 Rev. 4, section 3.3, describes emergency mitigation as the procedure "to temporarily mitigate vulnerabilities before a patch is available", and section 3.5.3 adds that "Schedules should be set and enforced for both patch deployment and mitigation removal." In this guide a mitigation lowers the chance of exploitation without fixing the flaw. Every row below comes from a vendor advisory or a record read on 19 Nov 2026, and the vendor is named where the claim is the vendor's. The verification column is the check to run after the change, from the device and from outside. The owner column names the role that makes the change and signs the check. | Type | Real example | What it costs | How to verify it took | Owner | | --- | --- | --- | --- | --- | | Restrict the interface | PAN-OS CVE-2026-0322, 14 Oct: limit the management interface to a jump box (Palo Alto advisory CVE-2026-0322) | Admins and scripts must come through that host | Read Device > Setup > Interfaces > Management Interface Settings > Permitted IP Addresses. Test the management address from an untrusted network and expect no answer | Network owner | | Disable the feature | FortiMail CVE-2026-104286, 1 Oct: turn off the IBE service (Fortinet) | IBE secure mail stops; recipients cannot open or reply through the portal | Read Encryption > IBE > IBE Service in the GUI, or the status line under config system encryption ibe in the CLI | Messaging service owner | | Remove a permission | ScreenConnect CVE-2026-84869, 3 Sep interim advisory (the bulletin and 26.6.5 followed on 8 Sep): deselect TransferFiles in every role (ConnectWise) | Technicians cannot transfer files in sessions | Open Administration > Security > Roles and check each session group in each role; the permission shows in bold when set | Remote support service owner | | Remove a setting | Cisco ASA and FTD CVE-2026-76418, 28 Oct: remove secondary authentication (Cisco) | The VPN loses its second credential, which Cisco says lowers its security posture | show running-config tunnel-group must show no secondary-authentication-server-group in any group, including DefaultWEBVPNGroup | Network security owner | | Filter at a proxy or WAF | Atlassian CVE-2026-21589, 5 Oct: block URLs with two dots beside a slash, backslash or double colon (Atlassian) | Rule upkeep and false positives; on Confluence, Jira and the others a restart of every node | From outside, request a harmless URL carrying an encoded ../ segment and confirm the proxy or the rewrite rule answers, not the application | Platform team that runs the proxy | | Take it off the internet | Atlassian CVE-2026-21589: remove the instance from the internet until patched (Atlassian); PHP CVE-2026-109618: CISA says take unpatchable systems offline | External users and integrations lose access | From a network outside yours, confirm the address no longer answers on its port | Network owner, service owner informed | | Limit the network path | Tomcat CVE-2026-34486: restrict the cluster receiver port to the other cluster nodes (Tomcat clustering guide) | A node added later needs a new rule | ss -ltnp shows the receiver (default port 4000, up to 4100 with autoBind). From a non-member host, a connection to it must fail | Application platform owner | | Set a secret | Artifactory CVE-2026-82329, 28 Aug: add your own additionalJoinKeys value (JFrog) | A restart of the Access service; the new key must be kept as secret as the join key | Confirm the entry in system.yaml, or JF_SHARED_SECURITY_ADDITIONALJOINKEYS in containers, and that the Access service restarted after the edit | Artifact repository service owner | | Change configuration | OpenSSH CVE-2026-123158, 5 Nov: PasswordAuthentication no, set in sshd_config including inside Match blocks (OpenSSH, Alpine) | Nobody can log in with a password; existing sessions keep running through the restart | sshd -t, restart sshd, then sshd -T \| grep -i passwordauthentication must print no. Test key login in a second session first | OS owner for the host | | Block a kernel module | Linux CVE-2026-46300: stop esp4 and esp6 from loading (Red Hat) | IPsec stops after the next reboot | lsmod \| grep -E 'esp4\|esp6' prints nothing, and /etc/modprobe.d/dirtyfrag.conf holds the install lines. Check lsmod first: if neither is loaded, IPsec is not in use | OS owner for the host | ## Some mitigations are partial, and the vendor says so PHP's advisory for PHP2Shell, CVE-2026-109618, lists memory_limit=-1 as an interim step and says it is not complete, because a failed mmap() can still reach the vulnerable path. PHP calls the upgrade the only complete mitigation. Verify the value from inside a worker, not from php.ini: ini_get('memory_limit') in a private test script must return -1, because a pool, directory or application setting can restore a finite limit. Palo Alto lists no workaround for CVE-2026-0322. The jump-box restriction is hardening: the flaw stays and the rating moves from CVSS 4.0 8.6 to 8.5, because the attack vector becomes adjacent. Palo Alto calls its Threat Prevention signatures, Threat IDs 510046 and 510047, limited coverage: they work only when management traffic is routed through a data-plane interface with certificate replacement and decryption. Cisco prefers the upgrade to its ASA mitigation and says the mitigation was proven in a test environment, so check it in yours. F5 lists one workaround for CVE-2026-94127, an iRule available on request, and does not say what it changes; test it on a non-production unit. A third-party note is not a vendor claim. Medcrypt reports that MaxAuthTries 1 keeps password login working while blocking its reproduction of the OpenSSH bypass and that MaxAuthTries 2 does not. It is a stopgap, and a user who mistypes a password once is disconnected. Some advisories offer none. Broadcom for ESX CVE-2026-41709 and Citrix for CVE-2026-19490 list no workaround, so the patch is the only fix they offer. ## Worked case: OpenSSH on Alpine, public events and a constructed team's hours All times are UTC. The public events are real. The team is constructed to show the order of work: it is the invented fleet of [where a new CVE hits you](/guides/where-a-new-cve-hits-you/), where checks cut 120 Linux hosts to nine in scope, four reachable from the internet and five internal. The mitigation clock starts when the fix goes public, not when a CVE ID exists. The ID, CVE-2026-123158, was published at 15:46 UTC on 5 Nov, 6 hours 46 minutes after the fix. A team that waited for the ID to start would have lost that time. | When | Hours after the fix | Public event | Constructed team's step | Owner | | --- | --- | --- | --- | --- | | Thu 5 Nov, 09:00 | 0 | OpenSSH fix pushed to the public repository at 09:00; 10.6p2 released at 09:01; Alpine publishes its advisory and fixed packages | Not yet seen | None | | Thu 5 Nov, 11:00 | 2 h | | Analyst reads the Alpine post, opens the decision log and declares the event: the fix is public and the flaw is an authentication bypass | Analyst on duty, then event lead | | Thu 5 Nov, 11:00 to 12:30 | 2 h to 3 h 30 min | | Scope: the five checks run across the fleet; nine hosts in scope; exposure sorted into four internet and five internal | Platform lead | | Thu 5 Nov, 12:30 to 13:15 | 3 h 30 min to 4 h 15 min | | Mitigate: PasswordAuthentication no on the four internet-facing hosts, key login tested in a second session first | Platform lead, change approver informed | | Thu 5 Nov, 13:16 | 4 h 16 min | Exploit code appears on GitHub (GreyNoise timeline, 13:16:55 UTC) | Verify: sshd -T on the four hosts prints no. Board brief 1 says four of four internet hosts mitigated, five internal in progress | Platform lead; scribe logs | | Thu 5 Nov, 14:30 | 5 h 30 min | | All nine hosts mitigated and verified; costs logged with an expiry of 7 Nov, 12:00 | Platform lead | | Thu 5 Nov, 15:46 | 6 h 46 min | CVE-2026-123158 published (GreyNoise timeline, 15:46:25 UTC) | The CVE ID is added to the log and the brief | Scribe | | Fri 6 Nov, 09:00 | 24 h | | Patch: apk upgrade -U -a on one canary host, then the four internet hosts, then sshd restarted with rc-service sshd restart, because Alpine says the package upgrade alone does not replace the running daemon; apk info -v openssh-server confirms the fixed package | Platform lead | | Fri 6 Nov, 14:00 | 29 h | | Patch the five internal hosts and verify | Platform lead | | Fri 6 Nov, 16:00 | 31 h | | Remove: PasswordAuthentication returns to the baseline where password login is needed; removal logged after apk info -v openssh-server confirmed the fixed package | Platform lead | | Sat 7 Nov, 00:46 | 39 h 46 min | GreyNoise sees the first successful exploitation on one of its sensors | All nine hosts patched and the mitigation closed 8 h 46 min earlier | Event lead closes the event | | Mon 9 Nov | About 4 days | CISA adds the CVE to KEV, due 12 Nov | Nothing open | None | ## Give every mitigation an owner, a review time and a removal condition The owner is a name, taken from the role in the table above. The review time is a clock time, no later than the next vendor update, since vendors revise advisories: Fortinet revised FG-IR-26-175 on 5 and 7 Oct 2026, and SonicWall updated its workaround on 9 Nov 2026. The removal condition is the fix, verified by version check or rescan. Tell support about the user-visible cost before it ships. Run a new WAF rule in log-only mode for an hour on a low-risk group if the exposure allows. If exploitation is active and the service is exposed, enforce first and loosen later. One mitigation must be in place before the patch, not after. WordPress 7.1.4's updater still creates a predictable folder under wp-content/upgrade while it installs the fix, so deny web access to that directory first (nginx: location ^~ /wp-content/upgrade/ { deny all; }; Apache: Require all denied). WordPress.org names no workaround; this rule follows from the 7.1.4 code change, which adds an index.php to that folder. ## A mitigation does not remove access the attacker already has Mitigations close a path from now on. FortiMail's KEV entry marks forensic triage as Yes, so preserve the logs and a configuration backup before you change the appliance. Artifactory shows the limit most clearly: patching does not revoke issued tokens, and Wiz reports that attacker Groovy plugins and stolen Access signing keys survive an upgrade, so the join-key workaround cannot undo an intrusion from before it was set. Until the fix is verified and the hunt is done, the board line reads open, with the mitigation named and its expiry. ## Related pages - [How to tell where a new CVE hits you](/guides/where-a-new-cve-hits-you/) - [Zero day vulnerability defined: the sources, the tests and six 2026 cases](/guides/zero-day-vulnerability-defined/) - [The 72-hour post-incident review: template and a worked review of the BeyondTrust zero-day](/guides/72-hour-post-incident-review/) ## Sources - Fortinet PSIRT advisory FG-IR-26-175: https://fortiguard.fortinet.com/psirt/FG-IR-26-175 (read 19 Nov 2026) - FortiMail CLI reference, system encryption ibe: https://docs.fortinet.com/document/fortimail/7.6.0/cli-reference/6402/system-encryption-ibe (read 19 Nov 2026) - Palo Alto Networks advisory CVE-2026-0322: https://security.paloaltonetworks.com/CVE-2026-0322 (read 19 Nov 2026) - Cisco advisory cisco-sa-asaftd-tunnel-dos-s6mgkb75: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-asaftd-tunnel-dos-s6mgkb75 (read 19 Nov 2026) - Atlassian advisory CVE-2026-21589: https://confluence.atlassian.com/security/cve-2026-21589-arbitrary-file-access-vulnerability-impacts-multiple-products-1870495748.html (read 19 Nov 2026) - ConnectWise ScreenConnect security bulletin, 8 Sep 2026: https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin (read 19 Nov 2026) - ConnectWise advisories page: https://www.connectwise.com/company/trust/advisories (read 19 Nov 2026) - JFrog security advisories, CVE-2026-82329 section and workaround: https://docs.jfrog.com/releases/docs/jfrog-security-advisories (read 19 Nov 2026) - F5 advisory K000162605: https://my.f5.com/manage/s/article/K000162605 (read 19 Nov 2026) - PHP security advisory GHSA-w8hw-w9fx-j79w: https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 19 Nov 2026) - CISA alert, Critical Vulnerabilities in PHP Require Immediate Patching: https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching (read 19 Nov 2026) - Alpine Linux advisory, OpenSSH password authentication bypass, 5 Nov 2026: https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/ (read 19 Nov 2026) - OpenSSH 10.6p2 release notes: https://www.openssh.org/releasenotes.html#10.6p2 (read 19 Nov 2026) - Medcrypt, independent reproduction and mitigation guidance for CVE-2026-123158: https://www.medcrypt.com/blogs/asterisk-hole-cve-2026-123158-when-a-locked-account-is-the-way-in (read 19 Nov 2026) - GreyNoise event page and timeline for CVE-2026-123158: https://sshole.events.greynoise.io/ (read 19 Nov 2026) - Apache Tomcat 11 clustering guide: https://tomcat.apache.org/tomcat-11.0-doc/cluster-howto.html (read 19 Nov 2026) - Red Hat security bulletin RHSB-2026-003: https://access.redhat.com/security/vulnerabilities/RHSB-2026-003 (read 19 Nov 2026) - Broadcom security advisory VMSA-2026-0006: https://support.broadcom.com/web/ecx/support-content-notification/-/external/content/SecurityAdvisories/0/38017 (read 19 Nov 2026) - WordPress 7.1.4 security release note: https://wordpress.org/news/2026/10/wordpress-7-1-4 (read 19 Nov 2026) - WordPress core source, 7.1.4 tag, class-wp-upgrader.php: https://raw.githubusercontent.com/WordPress/wordpress-develop/7.1.4/src/wp-admin/includes/class-wp-upgrader.php (read 19 Nov 2026) - Wiz Research, Artifactory exploitation report, 10 Sep 2026: https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201 (read 19 Nov 2026) - SonicWall PSIRT advisory SNWLID-2026-0014: https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014 (read 19 Nov 2026) - Citrix security bulletin CTX696939: https://support.citrix.com/external/article/CTX696939/netscaler-adc-and-netscaler-gateway-secu.html (read 19 Nov 2026) - CISA KEV feed, catalog version 2026.11.18: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026) - CISA KEV entry for CVE-2026-104286: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286 (read 19 Nov 2026) - NVD record for CVE-2026-123158: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026) - NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning: https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-40r4.pdf (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/mitigations-that-buy-time/ --- # To tell where a new CVE hits you, run the version, configuration and exposure checks in that order Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** To tell where a new CVE hits you, check three things in order: the version of every copy of the product, the configuration the advisory requires, and whether an attacker can reach the asset. Each has a command. For CVE-2026-123158, five configuration checks cut an example fleet of 120 hosts to nine. ## Read the version where the running software reports it Compare the reading with the fixed build in the advisory, not with a banner or a package name. Distribution packages carry backports under older version strings, so for OpenSSH and the Linux kernel read the package changelog or the vendor errata table. | Product and CVE | Command or page | Compare with | | --- | --- | --- | | PHP, CVE-2026-109618 | php-fpm -v, php-cgi -v, or php -v from the mod_php build, on every host and image | 8.2.35, 8.3.36, 8.4.27, 8.5.12 | | Node.js, CVE-2026-111497 | node --version on hosts, CI runners and inside images; libbrotli package version where Node links the system library | 22.23.4, 24.22.1, 26.12.1 | | fastjson, CVE-2026-16723 | mvn dependency:tree -Dincludes=com.alibaba:fastjson, or search the SBOM | Affected 1.2.68 to 1.2.83; fixed 1.2.84 | | OpenSSH, CVE-2026-123158 | ssh -V; on Alpine apk info -v openssh-server | 10.6p2 or the Alpine package table | | Linux kernel, CVE-2026-46300 | uname -r; on RPM systems rpm -q --changelog kernel \| grep CVE-2026-46300 | Distribution errata; the version number alone can mislead | | Cisco ASA and FTD, CVE-2026-76418 | show version | First fixed release for the train, such as ASA 9.20.4.50 or FTD 7.4.9 | | F5 BIG-IP, CVE-2026-94127 | tmsh show sys version and tmsh show sys software status, on each device in a failover pair | Hotfix-BIGIP-17.1.3.5.0.41.14-ENG for 17.1.x, Hotfix-BIGIP-17.5.1.9.0.160.12-ENG for 17.5.x, Hotfix-BIGIP-21.1.0.2.0.30.22-ENG for 21.1.0 | | PAN-OS, CVE-2026-0322 | show system info, sw-version line, on every firewall and on Panorama | 12.2.4; 12.1.4-h12, 12.1.7-h8 or 12.1.10-h2; 11.2.4-h22, 11.2.7-h21, 11.2.10-h15, 11.2.13-h3 or 11.2.16; 11.1.4-h37, 11.1.6-h39, 11.1.7-h11, 11.1.10-h35, 11.1.13-h15 or 11.1.16-h3; 10.2.7-h38, 10.2.10-h41, 10.2.13-h26, 10.2.16-h11 or 10.2.18-h12. Use the entry for your sub-train | | Artifactory, CVE-2026-82329 | GET /artifactory/api/system/version on every self-hosted instance | 7.111.21, 7.117.28, 7.125.20, 7.133.29, 7.146.38, 7.161.20 | | VMware ESX, CVE-2026-41709 | esxcli system version get | Build 25370933 (9.1), 25595025 (9.0), 25429389 (8.0) | | WordPress, CVE-2026-119977 | wp core version | 7.1.4, 7.0.8, 6.9.11 and so on down to 4.7.39 for older branches | | Windows, CVE-2026-68820 | winver, or [System.Environment]::OSVersion.Version | Fixed builds of 11 Aug 2026. Windows 11 24H2 and Server 2025 share the 26100 build number but are serviced by different packages, so their revisions differ (Microsoft CVRF): 11 24H2 10.0.26100.9106 or 9168; Server 2025 10.0.26100.33222 or 33296. Others: Server 2012 6.2.9200.26280; Server 2012 R2 6.3.9600.23338; 10 1607 and Server 2016 10.0.14393.9418; 10 1809 and Server 2019 10.0.17763.9121; 10 21H2 10.0.19044.7663; 10 22H2 10.0.19045.7663; Server 2022 10.0.20348.5440 or 5499; 11 23H2 10.0.22631.7517; 11 25H2 10.0.26200.9106 or 9168; 11 26H1 10.0.28000.2704 | ## Find the copies you did not install Search for the component, not only the product. The Node.js upgrade for CVE-2026-111497 fixes only the copy Node.js bundles. Every other copy of Brotli needs its own: Google Brotli 1.0.10, 1.1.1 or 1.2.1, the distribution package, Bun 1.4.4, PyPI Brotli 1.2.1. The PHP advisory covers PHP embedded in appliances and container base images as much as PHP you installed. To sweep images for a runtime, loop over them: for i in $(docker images --format '{{.Repository}}:{{.Tag}}'); do echo "$i $(docker run --rm --entrypoint node $i --version 2>/dev/null)"; done. Swap node for php-fpm or php. For libraries, search the dependency tree or the SBOM, and on hosts run ldconfig -p | grep -i brotli and pip list | grep -i brotli. For Java, the fastjson advisory's trigger is a Spring Boot executable fat-jar run with java -jar. Plain jar, uber-jar and WAR deployments do not meet it, so list how each service is packaged. ScreenConnect is the reverse case: servers are not affected, clients are. ConnectWise says not to rely on the console version alone, because updating the console does not update technician clients. ## Check the configuration the advisory requires A condition removes hosts from the list only if you test it on each host and keep the output. These are the commands for conditions in the 2026 records. | Condition | Command or page | In scope when | | --- | --- | --- | | OpenSSH CVE-2026-123158: musl without PAM | ldd $(command -v sshd) \| grep -iE 'musl\|pam'; sshd -T \| grep -i usepam | libc.musl appears, no PAM library, or UsePAM is no | | Password login on | sshd -T \| grep -iE 'passwordauthentication\|permitrootlogin'; also read Match blocks in /etc/ssh/sshd_config.d/ | passwordauthentication yes | | An account with no password, and one with a hash | awk -F: '$2=="*"{print $1}' /etc/shadow, then awk -F: '$2 ~ /^\$[56]\$/{print $1}' /etc/shadow | Both commands print at least one name. Accounts locked with ! do not count | | Cisco CVE-2026-76418: SSL VPN on | show running-config webvpn | A line enable | | Secondary authentication in scope | show running-config tunnel-group | secondary-authentication-server-group in any group, including DefaultWEBVPNGroup, even when tunnel-group-list is disabled | | F5 CVE-2026-94127: APM provisioned | tmsh list sys provision apm | A level other than none | | APM is an OAuth authorization server on a virtual server | tmsh list ltm virtual all-properties; GUI Access > Federation > OAuth Authorization Server | A virtual server with an access profile and an OAuth profile; an empty page means not in scope | | SonicWall CVE-2026-19868: SAML and an alternate server | SonicOS SSLVPN user authentication settings; Device > Users > Settings > Authentication > Configure RADIUS, LDAP or TACACS+ > Settings | SAML is the SSLVPN login method and any listed server is not disabled | | Artifactory CVE-2026-82329: default join key | Look under shared, security in system.yaml for additionalJoinKeys; in containers read JF_SHARED_SECURITY_ADDITIONALJOINKEYS | No additional join key set | | Tomcat CVE-2026-34486: clustering with EncryptInterceptor | version.sh; grep -nE 'Cluster\|EncryptInterceptor' conf/server.xml | Version exactly 11.0.20, 10.1.53 or 9.0.116, plus both strings in server.xml | | WordPress CVE-2026-119977: upgrade runs | wp config get WP_AUTO_UPDATE_CORE; Dashboard > Updates | Updates run automatically, or an admin or WP-CLI upgrade is planned | | Staging folder is public | curl -sI https://example.test/wp-content/upgrade/ from outside; ls -la wp-content/upgrade/ on the host | 200 or a listing means reachable; 403 or 404 means blocked. A folder named after the package with no random suffix is the unfixed behavior | ## Check the exposure from the attacker's side Sort matches into four groups: reachable from the internet, reachable from partner networks, reachable from the internal network, and isolated. Record the source of each answer: an external scan, a firewall rule export or an architecture note. Exposure from memory is the usual source of a missed host. Run the checks that need a vantage point from outside: curl -sI against the URL, the VPN login page of the ASA or SonicWall tested from an untrusted network, ss -ltnp | grep sshd on the host with the security-group rules beside it, and for Tomcat clusters ss -ltnp for ports 4000 to 4100. On PAN-OS, Permitted IP Addresses and every Network > Network Profiles > Interface Mgmt profile that allows HTTPS or SSH show what the management interface admits. On BIG-IP, the flaw is on the data plane, so the virtual server address and the self IPs that expose it decide exposure, not the management port. Do not drop the internal group. Some bugs need an attacker who already runs code, and CISA's KEV entry for the Linux flaw CVE-2026-46300 says it was chained with the two PHP2Shell CVEs to implant web shells and run commands as root. Windows CVE-2026-68820 needs a local user. Artifactory CVE-2026-42018 hands out a low-privilege token that CVE-2026-42016 turns into an admin token. For these, the exposure question is whether untrusted code or an untrusted token can reach the host. ## Who runs each check and what to keep as evidence The owner and the record turn a check into evidence a reviewer can use. Failures are listed once, by check. | Check | Run by | Output kept as evidence | How it fails | | --- | --- | --- | --- | | Version | Platform team, through configuration management or the product's own page | The command output per host, with host name and UTC time | A backport: the version string is below the fixed one but the package is patched. Keep the changelog line or errata entry beside the reading | | Copies you did not install | Application owners, with the build team | Image digests and the SBOM query result | Images not scanned: the sweep covers running containers but not stopped ones, base images in the registry or CI runners. Name what the sweep covered | | Configuration | Platform team | The configuration command output per host, including Match blocks and drop-in files | One member of a failover pair, or one drop-in file, is read and the other is not | | Exposure | Network owner, from outside the network | The external scan or request output and the firewall rule export, each dated | Stale exposure data: the scan or export is older than the last firewall or security-group change. Rerun it, and print its date in the record | ## Worked scoping: five checks narrow 120 hosts to nine for CVE-2026-123158, and a sixth sorts them by exposure The counts below are an invented fleet of 120 Linux hosts to show the arithmetic; the conditions and commands come from the record. Each row runs on the survivors of the row before, and each excluded host gets a one-line reason in the record. | Check | Hosts left | | --- | --- | | Start: Linux hosts running sshd | 120 | | sshd links musl | 31 | | No PAM, or UsePAM no | 26 | | PasswordAuthentication yes | 19 | | At least one account with * as its password field | 12 | | At least one other account with a $5$ or $6$ hash | 9 | | Port 22 reachable from an untrusted network | 4 internet, 5 internal | ## When the usual answer is wrong A version number below the fixed one does not always mean exposed. Alpine's fixed OpenSSH package for 3.22 is 10.0_p1-r11 and for 3.21 is 9.9_p2-r1. Both report a version far below 10.6p2 and are patched for CVE-2026-123158, so compare the package release with the vendor table, not with the upstream number. A version at or above a fixed-looking number does not always mean safe. OpenSSH 10.6 and 10.6p1 are affected; only 10.6p2 carries the fix. The same holds in the other direction for configuration: a Cisco ASA on a vulnerable release without secondary authentication on any tunnel group, including DefaultWEBVPNGroup, is outside CVE-2026-76418, and a ScreenConnect server at any version tells you nothing about the clients. ## Repeat the match when the advisory changes Advisories grow. Fortinet revised FG-IR-26-175 on 5 Oct 2026 and again on 7 Oct, when it named FortiMail 7.2.10 as an upcoming fix. SonicWall's version 1.1 of 9 Nov updated its workaround. WordPress published 7.1.5 on 18 Nov, after the 7.1.4 release of 28 Oct. Keep the advisory URL and its read time in the record so the second read is a diff against the first, and rerun the match at 24 hours and after the patch. Log each gap the inventory missed. A recurring one, such as components bundled in vendor appliances, belongs in the supplier questionnaire as a request for a component list. The tools page shows where 14 vendors say a new CVE lands. ## Related pages - [The 72-hour post-incident review: template and a worked review of the BeyondTrust zero-day](/guides/72-hour-post-incident-review/) - [Mitigations that buy time before the patch](/guides/mitigations-that-buy-time/) - [Compromise assessment after patching a critical flaw](/guides/compromise-assessment-after-patching/) ## Sources - PHP security advisory GHSA-w8hw-w9fx-j79w: https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 19 Nov 2026) - Node.js security releases, 21 Oct 2026: https://nodejs.org/en/blog/vulnerability/october-2026-security-releases (read 19 Nov 2026) - fastjson2 wiki, security advisory for fastjson 1.2.68 to 1.2.83: https://github.com/alibaba/fastjson2/wiki/Security-Advisory:-Remote-Code-Execution-in-fastjson-1.2.68%E2%80%931.2.83 (read 19 Nov 2026) - Alpine Linux advisory, OpenSSH password authentication bypass, 5 Nov 2026: https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/ (read 19 Nov 2026) - OpenSSH 10.6p2 release notes: https://www.openssh.org/releasenotes.html#10.6p2 (read 19 Nov 2026) - Red Hat CVE page for CVE-2026-46300: https://access.redhat.com/security/cve/CVE-2026-46300 (read 19 Nov 2026) - Cisco advisory cisco-sa-asaftd-tunnel-dos-s6mgkb75: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-asaftd-tunnel-dos-s6mgkb75 (read 19 Nov 2026) - F5 advisory K000162605: https://my.f5.com/manage/s/article/K000162605 (read 19 Nov 2026) - SonicWall PSIRT advisory SNWLID-2026-0014: https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014 (read 19 Nov 2026) - Palo Alto Networks advisory CVE-2026-0322: https://security.paloaltonetworks.com/CVE-2026-0322 (read 19 Nov 2026) - JFrog security advisories, CVE-2026-82329 section and workaround: https://docs.jfrog.com/releases/docs/jfrog-security-advisories (read 19 Nov 2026) - JFrog documentation, Get Artifactory Version API: https://docs.jfrog.com/administration/reference/getartifactoryversion (read 19 Nov 2026) - Broadcom security advisory VMSA-2026-0006: https://support.broadcom.com/web/ecx/support-content-notification/-/external/content/SecurityAdvisories/0/38017 (read 19 Nov 2026) - WordPress 7.1.4 security release note: https://wordpress.org/news/2026/10/wordpress-7-1-4 (read 19 Nov 2026) - WordPress core source, 7.1.4 tag, class-wp-upgrader.php: https://raw.githubusercontent.com/WordPress/wordpress-develop/7.1.4/src/wp-admin/includes/class-wp-upgrader.php (read 19 Nov 2026) - Microsoft MSRC CVRF, August 2026 release: https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug (read 19 Nov 2026) - Microsoft MSRC record for CVE-2026-68820: https://api.msrc.microsoft.com/sug/v2.0/en-US/vulnerability/CVE-2026-68820 (read 19 Nov 2026) - Apache Tomcat 11 clustering guide: https://tomcat.apache.org/tomcat-11.0-doc/cluster-howto.html (read 19 Nov 2026) - Apache Tomcat 11 security page: https://tomcat.apache.org/security-11.html (read 19 Nov 2026) - Fortinet PSIRT advisory FG-IR-26-175: https://fortiguard.fortinet.com/psirt/FG-IR-26-175 (read 19 Nov 2026) - ConnectWise ScreenConnect security bulletin, 8 Sep 2026: https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin (read 19 Nov 2026) - GreyNoise event page and timeline for CVE-2026-123158: https://sshole.events.greynoise.io/ (read 19 Nov 2026) - CISA KEV entry for CVE-2026-46300: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-46300 (read 19 Nov 2026) - NVD record for CVE-2026-123158: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026) - NVD record for CVE-2026-42016: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-42016 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/where-a-new-cve-hits-you/ --- # The 72-hour post-incident review is a 90-minute meeting with a fixed agenda, a timing table and a changes list Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** Held on day three or four, the review takes 90 minutes, one facilitator who held no role, and a record of eight timestamps. The template below runs the agenda, then a worked review of BeyondTrust CVE-2026-102764 shows how 47 days of exposure before the 15 Oct 2026 advisory changes what the review asks. ## What NIST says the review is for NIST SP 800-61r3 treats the review as part of improvement, not a ritual at the end. Its recovery guidance says to "Prepare an after-action report that documents the incident itself, the response and recovery actions taken, and lessons learned." It also notes that "Improvements are often identified when creating follow-up reports for incidents or holding 'lessons learned' meetings", and that lessons "should often be shared as soon as they are identified, not delayed until after recovery concludes." That last quotation is why the review is at 72 hours and not at the end: memory of who decided what fades in days. ## The template Copy this into the incident record. The security lead fills the header and the timestamps before the meeting. The scribe, named in advance, fills the rest during it. - Header: incident ID; CVE and product; advisory link and the time it was read; review date and start time; facilitator (held no role in the incident); scribe; attendees by role (security lead, platform engineering, application owners, communications, one executive). - Timestamps, each with its source log: advisory published; advisory first read by your team; exposed-asset list issued; first control in place; first patched ring; last in-scope asset fixed; verification rescan clean; compromise assessment closed. - Question 1, when did you know? What was the gap between publication and your first read, and which feed or person caught it? - Question 2, what was decided and by whom? For each choice (mitigate, patch, accept) record the approver, the time and any expiry. - Question 3, how long did each step take? Fill the four measures from the timestamps; name the step that took longest. - Question 4, what was missed? Assets absent from inventory, bundled copies, hosts owned by other teams. - Question 5, what did the tools say? For the inventory, scanner, WAF, endpoint tool or any triage agent: right, late or wrong, with one example each. - Question 6, did anyone get in first? The window searched, the logs searched, the finding, and the closing time. - Changes: at most five; each with an owner, a due date and a check date. The next review reads this list first. - Publication: a one-page internal summary with the four measures, the findings and the changes. ## Agenda and who records what Send the timestamps one day ahead so the meeting is for decisions. | Minutes | Item | Led by | Record left | | --- | --- | --- | --- | | 0 to 10 | Read the timestamps and the four measures aloud | Facilitator | Confirmed or corrected timestamps | | 10 to 30 | Questions 1 and 2: knowing and deciding | Security lead | Decision log with approvers | | 30 to 50 | Questions 3 and 4: elapsed time and misses | Platform engineering | Longest step; missed-asset list | | 50 to 65 | Question 5: what the tools said | Application owners | Tool findings, one per tool | | 65 to 75 | Question 6: did anyone get in first | Security lead | Assessment window and finding | | 75 to 90 | Changes list, owners and dates | Facilitator | Five changes or fewer | ## The four measures Time to scope runs from advisory published to exposed-asset list issued. Time to mitigate runs from advisory published to the first control on exposed assets. Time to patch runs from advisory published to the last in-scope asset fixed. Time to verify runs from the last fix to a clean rescan. Four are enough; more invite arguments about definitions. ## The public dates for BeyondTrust CVE-2026-102764, advisory BT26-05 These dates are public and fix the clock for every customer with a self-hosted Remote Support appliance. BeyondTrust and Mandiant disclosed the bug on 15 Oct 2026; the vendor had released the BT26-05 patch on 8 Oct. CISA added it to its catalog on 16 Oct with a due date of 19 Oct. Mandiant dates first exploitation to 29 Aug. | Interval | From | To | Elapsed | | --- | --- | --- | --- | | Exposure before advisory | 29 Aug 2026, first exploitation | 15 Oct 2026, advisory | 47 days | | Exploitation to detection in the hosted instance | 29 Aug 2026 | 28 Sep 2026 | 30 days | | Patch released before advisory | 8 Oct 2026 | 15 Oct 2026 | 7 days | | Advisory to catalog entry | 15 Oct 2026 | 16 Oct 2026 | 1 day | | Advisory to catalog due date | 15 Oct 2026 | 19 Oct 2026 | 4 days | ## A worked review: Northgate Freight, an invented company with two self-hosted appliances Northgate Freight is invented for this example, as are all of its times (UTC). It runs a production Remote Support appliance, RS-1, and a disaster-recovery appliance, RS-2, both on 26.2.3. The advisory time is taken as 08:00 on Thu 15 Oct 2026, the public date. The review is held on Mon 19 Oct at 10:00, day four. The security lead fills this table before the meeting. The four measures follow from it. | Timestamp | Time (UTC) | Source log | | --- | --- | --- | | Advisory published | Thu 15 Oct, 08:00 | Vendor advisory (assumed time) | | Advisory first read by the team | Thu 15 Oct, 09:10 | On-call channel post | | Exposed-asset list issued (RS-1 only) | Thu 15 Oct, 14:00 | Ticket BT-1 | | First control in place: /login and /appliance on RS-1 limited to office and VPN ranges | Thu 15 Oct, 16:30 | Firewall change record | | First patched ring: RS-1 | Fri 16 Oct, 02:00 | Appliance update log | | RS-2 found and added to the list | Fri 16 Oct, 09:00 | Network team firewall export | | Last in-scope asset fixed: RS-2 | Sat 17 Oct, 02:00 | Appliance update log | | Verification rescan clean | Sat 17 Oct, 10:00 | Scanner report | | Compromise assessment closed | Open; due Tue 20 Oct, 17:00 | Assessment ticket | ## The four measures and the answers to the six questions Time to scope: complete list at Fri 16 Oct 09:00, 25 hours after publication (the first list, at 6 hours, missed RS-2). Time to mitigate: 8 hours 30 minutes. Time to patch: 42 hours, finished Sat 17 Oct, two days before CISA's due date of 19 Oct. Time to verify: 8 hours. Question 1, when did you know? The team read the advisory 1 hour 10 minutes after publication because a person watching the vendor feed posted it. The patch had existed since 8 Oct, seven days before the advisory, and Northgate had no step that picks up a vendor patch released without an advisory. Question 2, what was decided and by whom? The security lead approved the access restriction on RS-1 at 16:30, with an expiry at patching. The change approver approved patching RS-1 on 15 Oct at 17:00 and RS-2 on 16 Oct at 09:30, half an hour after it was found. A snapshot with memory was taken before RS-1 was patched, because BeyondTrust says upgrading alone does not remove an earlier compromise and Mandiant advises treating any internet-reachable unpatched appliance as possibly compromised. Question 3, how long did each step take? The longest step was 24 hours, from RS-1 patched (Fri 16 Oct 02:00) to RS-2 patched (Sat 17 Oct 02:00), because RS-2 was not on the first list and waited for the next window. The other steps: 1 hour 10 minutes to read, 4 hours 50 minutes to list, 2 hours 30 minutes to the first control, 9 hours 30 minutes from control to the first patch, 8 hours to verify. Question 4, what was missed? RS-2, absent from the inventory. Appliance syslog was not forwarded off the box, and Mandiant reports the actor deleted local logs, among them /ns/log/auth.log and /home/bomgar/log/bt-secrets.log, so the appliances hold no reliable record for 29 Aug to 15 Oct. Question 5, what did the tools say? Inventory: wrong, it listed RS-1 only. Scanner: late, its check for this bug loaded on Fri 16 Oct at 06:00, after RS-1 was patched, so the Sat 17 Oct rescan was its first useful read. SIEM: blind, because no appliance events reached it. Question 6, did anyone get in first? The window runs from 29 Aug, 47 days before the advisory, not from 15 Oct. BeyondTrust found an intrusion in its own hosted instance on 28 Sep, 30 days after exploitation began, and contained it within 27 minutes of the finding, so exploitation before disclosure is documented. For Northgate the finding is open until Tue 20 Oct at 17:00, and the review records it that way. Changes, each with an owner and a due date: forward appliance syslog and session events to the SIEM in near real time (platform engineering, 30 Oct); keep /login and /appliance restricted to trusted networks (network team, done 15 Oct); allow outbound traffic from the appliances only to required destinations (network team, 30 Oct); add the vendor's patch notes feed to the on-call channel (security lead, 23 Oct); add RS-2 and every appliance to the inventory (IT operations, 23 Oct). ## When the usual review is the wrong one Exploitation before disclosure moves the start of every clock. Time to patch measured from the advisory hides 47 days of exposure, so for a zero-day the review must open with the window in Question 6. If the compromise assessment is still open at 72 hours, hold the review anyway, mark Question 6 as open with a closing time, and schedule a short second session. For an incident that runs for weeks, follow NIST and share each lesson as it appears instead of waiting for one meeting. A review that ends with changes lacking owners is worse than none, because it reads as finished. ## Related pages - [Compromise assessment after patching a critical flaw](/guides/compromise-assessment-after-patching/) - [How to tell where a new CVE hits you](/guides/where-a-new-cve-hits-you/) - [Zero day response plan: the complete template with owners, triggers and a drill](/guides/zero-day-response-plan/) ## Sources - NIST SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cyber Risk Management: https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-61r3.pdf (read 19 Nov 2026) - BeyondTrust advisory BT26-05: https://www.beyondtrust.com/trust-center/security-advisories/bt26-05 (read 19 Nov 2026) - BeyondTrust blog on BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - Mandiant and GTIG analysis of the BeyondTrust zero-day: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog entry for CVE-2026-102764: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-102764 (read 19 Nov 2026) - NVD record for CVE-2026-102764: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-102764 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/72-hour-post-incident-review/ --- # After patching, hunt the exposure window for what the patch leaves behind Guide, 19 Nov 2026, read 19 Nov 2026 **The short answer.** A patch closes the flaw and leaves whatever came through it. The worked hunt here covers JFrog Artifactory CVE-2026-82329: a window from 28 Aug 2026, join-endpoint requests answered with HTTP 201, unknown admin users, Groovy plugins, truncated logs, then rotation of the join key and tokens issued since 28 Aug. ## A compromise is a judgment on the evidence NIST's glossary, quoting CNSSI 4009-2022, defines a compromise as "a judgment, based on the preponderance of the evidence, that a disclosure of information to unauthorized persons or a violation of the security policy for a system in which unauthorized, intentional or unintentional disclosure, modification, destruction, or loss of an object has occurred." A compromise assessment is the work that produces that judgment after a patch: it searches the exposure window and writes what the evidence supports. ## The window opens at the advisory, and earlier if you ran the chained bugs JFrog and NVD published CVE-2026-82329 on 28 Aug 2026. Fastly saw no exploitation that day. It counted only single-digit probes on 29 and 30 Aug, then about 75,000 requests on 31 Aug, nearly 98% from offensive-security services. A public exploit appeared on 1 Sep, with about 171,000 attempts. The peak came on 2 Sep, at about 406,000. CISA added the bug to its catalog that day. Wiz confirmed successful exploitation by several actors from 1 to 8 Sep. Wiz also saw CVE-2026-42018 and CVE-2026-42016 chained on Artifactory from 15 Aug 2026. If an instance ran builds exposed to those two as well, start the window on 15 Aug. The example below assumes the instance was exposed only to CVE-2026-82329 and was upgraded on 3 Sep 2026 (an assumption for the walk-through, not a reported event). That gives a seven-day window, 28 Aug to 3 Sep. The window does not end on patch day. Fastly states that patching does not revoke tokens already issued, so an admin token minted on 2 Sep still works on 20 Sep. Search for use of tokens and accounts from 28 Aug until the day they are revoked. | Date (2026) | Public activity in the vendor and research reports | What the date tells your search | | --- | --- | --- | | 28 Aug | Advisory and NVD record published; Fastly sees no exploitation | Earliest day to search; also the date from which to revoke tokens | | 29 and 30 Aug | Single-digit probes (Fastly) | A 201 here would be an early, targeted hit, not noise | | 31 Aug | About 75,000 requests, nearly 98% from offensive-security services | Expect many rejected requests; only a 201 matters | | 1 Sep | Public exploit; about 171,000 attempts; Wiz's successful exploitation begins | Highest-value day for new admin users and plugins | | 2 Sep | About 406,000 attempts; CISA adds the bug to KEV | Peak volume; look for the same source repeating across many hosts | | 3 Sep | Fastly publishes its analysis; the example instance is upgraded | Copy evidence before the restart; see the next section | | 5 Sep | CISA due date for federal agencies | Reference point if your policy follows the catalog | | 8 Sep | Last day of Wiz's observed successful exploitation | Window for reported attacks ends; yours runs until tokens are revoked | ## Copy evidence off the host before the upgrade The upgrade restarts the service, and attackers may truncate request*.log files (Wiz), so a log copied late can be a log already cut. Before the upgrade, copy these to storage the Artifactory host cannot write to: - The access log of the reverse proxy or load balancer in front of Artifactory, for 28 Aug to the copy time. Edge logs survive an attacker who cleans the host. - Every request*.log under the Artifactory log directory, with file sizes and modification times noted. - A listing of $JFROG_HOME/var/etc/artifactory/plugins/ with timestamps. - The system.yaml file, to show whether additionalJoinKeys was set. - An export of the user list, including admin flags and any SSH keys attached to users. ## Six searches, each with an owner and a record Run them in this order. Log formats differ, so adjust the field positions to your proxy. The indicator values come from Fastly and Wiz, read 19 Nov 2026. Search 1, the entry call. Owner: security lead. In the edge log, find POST requests to /access/api/v1/registry/join, /access/api/v1/registry/join/router and the ?override=true form. Example: grep -E 'POST [^ ]*/access/api/v1/registry/join' edge-access.log | grep ' 201 '. The key ID (kid) sits in the header of the token in the POST body, so it shows only where bodies are captured (a WAF, a proxy with body logging, a packet capture). Fastly treats a 201 carrying the kid e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855, the SHA-256 of an empty string, as high-confidence evidence of compromise. A normal cluster peer signs with a real join key and shows a different kid. Wiz saw 200 or 201 with an admin-scoped token. Without body capture, any 200 or 201 on these paths from an address outside your own nodes needs an explanation. Record: the command, the log files and dates, the hit count. Search 2, source addresses and user agents. Owner: security lead. Compare sources with the sample high-volume addresses Fastly lists: 213.136.72[.]87, 89.117.49[.]147, 45.33.75[.]219, 23.92.20[.]11, 173.255.226[.]148, 195.154.157[.]234, 62.210.87[.]212, 45.79.190[.]169 and 163.172.132[.]245, and with the user agents jfrt/0/7.161.19, JFrog-Access/7.141.17 and JFrogArtifactory/7.146.25. Both are easy to change, so a miss proves little. Record: the list compared and its source. Search 3, follow-on calls. Owner: security lead. Look in the same logs for the calls Wiz lists: GET /api/system/configuration, PUT /api/security/users/ with an admin flag in customData, token minting at /access/api/v1/tokens, reads of /access/api/v1/system/security/join_key, and enumeration of users, repositories and tokens. Record: each call, time and source address. Search 4, accounts. Owner: application owner, who knows the real users. Export users and compare with the identity provider, including admin flags and attached SSH keys. Account names Wiz saw: Nxploited_ plus 3 characters, labadmin_ plus hex, 0xTerror, svc_ plus 8 characters, and look-alikes such as jfrog-distribution, jfrog-insight and repo-service. Record: each unexpected user with its creation time and the decision taken. Search 5, plugins and filters. Owner: platform engineering. Run find "$JFROG_HOME/var/etc/artifactory/plugins" -type f -newermt '2026-08-28' -ls and look for the Groovy plugins Wiz names: rce.groovy, cmd.groovy, jwtToken.groovy, uploadFile.groovy, JFrogImage.groovy, metrics.groovy and httpSession.groovy, which are run at /artifactory/api/plugins/execute/. A broken plugin API, or requests to registry/join, api/plugins or systemUpload rejected with HTTP 400, can mean an in-memory filter named traceFilter is installed (Wiz). The upgrade leaves plugins and filters in place; restarting the service from a clean state clears the filter, but a file on disk stays. Record: file names, timestamps and hashes. Search 6, log integrity. Owner: platform engineering. Run find -name 'request*.log' -size 0 -newermt '2026-08-28'. Empty or missing request logs are themselves an indicator (Wiz). Record: file sizes and times against the copies taken before the upgrade. ## The finding decides what you rotate and whether the incident plan starts Rotate the platform join key and revoke access tokens issued since 28 Aug 2026 in every case. That advice comes from Fastly and does not depend on the finding, because a clean search of logs that an attacker could have cut cannot rule out a token. Write the finding as what was searched, over which dates, with which result. | Finding | What the searches showed | Action | | --- | --- | --- | | No evidence of access | No 200 or 201 on the join paths from unknown sources; no unknown users, plugins or filters; request logs intact | Record the logs and dates searched. Rotate the join key and revoke tokens since 28 Aug. | | Attempts without success | Join requests from unknown sources, all rejected | As above, and restrict network access to Artifactory and the Access paths to trusted systems (Wiz). | | Compromise | A 201 with the empty-key ID, an unknown admin user, an unexplained plugin, or cut logs | Start the incident plan built on NIST SP 800-61r3. Rotate credentials stored in Artifactory, review artifacts published or changed since 28 Aug, and, if private.key was read, reset the token certificate. A patch does not revoke forged tokens. | ## Other products leave different traces The same method fits any edge product. Replace the six searches with the vendor's own indicators from its record, keep an owner and a record for each, and apply the same finding table. Where compromise may have reached the host itself, rebuilding is the vendor's route. Cisco strongly recommends re-imaging suspected ISE nodes, and Mandiant's guidance for BeyondTrust is to rebuild from a clean patched image rather than clean in place. | Product and bug | Exposure window start | Indicator named in the record | | --- | --- | --- | | BeyondTrust Remote Support, CVE-2026-102764 | 29 Aug 2026, 47 days before the advisory of 15 Oct 2026 | Files mangoapi.php, /tmp/ingredi_ws and /usr/bin/dbus-authd; missing /ns/log/auth.log and /ns/log/http-access.log; user_changed events raising roles to Administrator | | Cisco ISE, CVE-2026-76460 | 16 Sep 2026, the advisory day; Cisco gives no earlier date | Suspicious usernames in the ise-kong access.log on every node; unexpected uploads from the node in firewall logs | ## Who owns each part, and where assessments fail The security lead owns the finding. Platform engineering exports the logs and snapshots. The application owner says what normal looks like, such as which addresses are real cluster peers. The record is the evidence folder, the six search results with commands and dates, and the written finding, kept with the incident record. Assessments usually fail in three ways. The logs were copied after the restart. The window was set from the patch date instead of the exposure date. Or a clean result was written as no compromise instead of what was searched. For Artifactory Cloud, JFrog fixed the service itself, so the self-hosted steps above do not apply. ## Related pages - [Zero day response plan: the complete template with owners, triggers and a drill](/guides/zero-day-response-plan/) - [The 72-hour post-incident review: template and a worked review of the BeyondTrust zero-day](/guides/72-hour-post-incident-review/) - [What to do when a zero day is disclosed: the first four hours, worked on PHP2Shell](/guides/what-to-do-when-a-zero-day-is-disclosed/) ## Sources - JFrog security advisories, CVE-2026-82329 section and workaround: https://docs.jfrog.com/releases/docs/jfrog-security-advisories (read 19 Nov 2026) - Fastly Security Research on CVE-2026-82329, 3 Sep 2026: https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity (read 19 Nov 2026) - Wiz Research on Artifactory exploitation, 10 Sep 2026, updated 17 Sep 2026: https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities catalog entry for CVE-2026-82329: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-82329 (read 19 Nov 2026) - NVD record for CVE-2026-82329: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-82329 (read 19 Nov 2026) - BeyondTrust blog on BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - Mandiant and GTIG analysis of the BeyondTrust zero-day: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - Cisco Security Advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - NIST SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cyber Risk Management: https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-61r3.pdf (read 19 Nov 2026) - NIST CSRC glossary, compromise (CNSSI 4009-2022): https://csrc.nist.gov/glossary/term/compromise (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/guides/compromise-assessment-after-patching/ --- # 20 terms a reader meets, each with a sourced example Glossary, read 19 Nov 2026 **The short answer.** Twenty terms from zero-day response, each defined in words quoted from NIST, CISA, FIRST or MITRE, or from the Clock rules, with a dated 2026 example such as the BeyondTrust flaw exploited on 29 Aug 2026, 47 days before its advisory. Every entry names its source and shows a read date of 19 Nov 2026. - [Attack surface](#attack-surface) - [Binding operational directive](#binding-operational-directive) - [Compensating control](#compensating-control) - [CVE identifier](#cve-identifier) - [CVSS](#cvss) - [CWE](#cwe) - [Disclosure Clock](#disclosure-clock) - [Disclosure date](#disclosure-date) - [Emergency directive](#emergency-directive) - [EPSS](#epss) - [Forensic triage](#forensic-triage) - [Incident](#incident) - [Indicator of compromise](#indicator-of-compromise) - [Known Exploited Vulnerabilities catalog](#kev-catalog) - [Mitigation](#mitigation) - [Patch management](#patch-management) - [Remediation](#remediation) - [Security advisory](#security-advisory) - [Software bill of materials](#software-bill-of-materials) - [Zero-day vulnerability](#zero-day-vulnerability) ## Attack surface NIST CSRC defines an attack surface as "The set of points on the boundary of a system, a system component, or an environment where an attacker can try to enter, cause an effect on, or extract data from, that system, component, or environment." Example: BeyondTrust advised on 15 Oct 2026, for CVE-2026-102764, that [self-hosted Remote Support appliances](/playbooks/remote-support-tool/) expose only the parts users need and restrict /login and /appliance to trusted networks, which shrinks the reachable points. (example from [BeyondTrust advisory BT26-05](https://www.beyondtrust.com/trust-center/security-advisories/bt26-05), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/attack_surface), read 19 Nov 2026) ## Binding operational directive CISA describes a binding operational directive as "a compulsory direction to federal, executive branch, departments, and agencies for purposes of safeguarding federal information and information systems." BOD 26-04, Prioritizing Security Updates Based on Risk, is the current directive on remediation timelines for known exploited vulnerabilities. Example: BOD 26-04, dated 10 Jun 2026, revokes BOD 22-01 and BOD 19-02. For CVE-2026-0322 its timeline began when CISA added the flaw to the catalog on 14 Oct 2026. (CVE-2026-0322 example from [CISA catalog entry](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322), read 19 Nov 2026) ([source](https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk), read 19 Nov 2026) ## Compensating control NIST CSRC defines a compensating security control as "A management, operational, and/or technical control (i.e., safeguard or countermeasure) employed by an organization in lieu of a recommended security control in the low, moderate, or high baselines that provides equivalent or comparable protection for an information system." Example: For Cisco ISE CVE-2026-76460, disclosed 16 Sep 2026, Cisco listed no workaround and offered infrastructure access control lists, which limit who reaches the API without fixing the flaw. (example from [Cisco advisory cisco-sa-ISE-ABP-VNSW7Tn5](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/compensating_security_control), read 19 Nov 2026) ## CVE identifier NIST CSRC gives Common Vulnerabilities and Exposures (CVE) as "A dictionary of common names for publicly known information system vulnerabilities." The identifier is a name only. Severity comes from CVSS, and exploitation status comes from sources such as the CISA catalog. Example: CVE-2026-109618 and CVE-2026-109619 are two names for PHP flaws disclosed on 21 Oct 2026, both fixed in PHP 8.2.35, 8.3.36, 8.4.27 and 8.5.12. (example from [PHP security advisory GHSA-w8hw-w9fx-j79w](https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/common_vulnerabilities_and_exposures), read 19 Nov 2026) ## CVSS The FIRST CVSS v4.0 specification says "The Base Score reflects the severity of a vulnerability according to its intrinsic characteristics which are constant over time and assumes the reasonable worst-case impact across different deployed environments." Threat and Environmental metrics then adjust the result for exploitation and for the local setting. Example: Palo Alto Networks rated PAN-OS CVE-2026-0322, disclosed 14 Oct 2026, 8.6 on CVSS 4.0, and 8.5 when the management interface sits behind a single jump box. (example from [Palo Alto Networks advisory CVE-2026-0322](https://security.paloaltonetworks.com/CVE-2026-0322), read 19 Nov 2026) ([source](https://www.first.org/cvss/v4-0/specification-document), read 19 Nov 2026) ## CWE The MITRE CWE FAQ defines a weakness as "a condition in a software, firmware, hardware, or service component that, under certain circumstances, could contribute to the introduction of vulnerabilities." One weakness type can sit behind many CVE records, so a CWE names the kind of flaw, not the single bug. Example: [FortiMail](/playbooks/email-gateway/) CVE-2026-104286, disclosed 1 Oct 2026, carries CWE-22, path traversal; the Fortinet advisory and the CISA catalog also list CWE-158, a null byte weakness. (example from [Fortinet advisory FG-IR-26-175](https://fortiguard.fortinet.com/psirt/FG-IR-26-175), read 19 Nov 2026) ([source](https://cwe.mitre.org/about/faq.html), read 19 Nov 2026) ## Disclosure Clock The Disclosure Clock is this site's dataset of days from disclosure to first dated exploitation for headline bugs of 2026. Disclosure is the vendor advisory date, or the NVD published date when none is stated. First exploitation is the earliest dated public report, else the CISA catalog date added, which is an upper bound on the first exploitation date. Example: Read 19 Nov 2026: 33 bugs have a dated exploitation, the median is 2 days, 24 were exploited within 7 days, 16 within 1 day and 12 on or before disclosure day. ([source](https://zerodayplaybook.com/#zp-clock), read 19 Nov 2026) ## Disclosure date The disclosure date is the day a vendor advisory made a vulnerability public, or the NVD published date when no advisory states one. The Disclosure Clock counts days to first exploitation from it. A researcher post or a CISA catalog listing does not set it. Vendors can revise advisories later. Example: WordPress published the 7.1.4 release note for CVE-2026-119977 on 28 Oct 2026 and NVD published the record on 31 Oct 2026, so the Clock uses 28 Oct, and exploitation on 3 Nov 2026 is day 6. (example from [WordPress 7.1.4 release note](https://wordpress.org/news/2026/10/wordpress-7-1-4) and the [NVD record](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-119977), read 19 Nov 2026) ([source](https://zerodayplaybook.com/#zp-clock), read 19 Nov 2026) ## Emergency directive CISA's page for ED 26-03 cites 44 U.S.C. section 3553(h), which lets the Secretary of Homeland Security "issue an emergency directive to the head of an agency to take any lawful action with respect to the operation of the information system" in response to a substantial threat. Each directive sets actions and deadlines. Example: ED 26-03 on Cisco SD-WAN, issued 25 Feb 2026, set an inventory deadline of 26 Feb, updates by 27 Feb, and reports to CISA on 26 Feb, 5 Mar, 12 Mar and 23 Mar 2026. ([source](https://www.cisa.gov/news-events/directives/v1-ed-26-03-mitigate-vulnerabilities-cisco-sd-wan-systems), read 19 Nov 2026) ## EPSS FIRST describes the Exploit Prediction Scoring System as "a data-driven machine-learning model that estimates the probability that a published CVE will be exploited in the wild in the next 30 days." It publishes a probability from 0 to 1 for every CVE each day. It forecasts; the CISA catalog records exploitation that has happened. Example: Cisco ISE CVE-2026-76460 was in the CISA catalog on its disclosure day, 16 Sep 2026, so for that flaw exploitation was a recorded fact rather than a 30 day forecast. (example from [CISA catalog entry](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460), read 19 Nov 2026) ([source](https://www.first.org/epss/), read 19 Nov 2026) ## Forensic triage For the shortest timeline, BOD 26-04 says the agency must complete remediation or mitigation within three days and "carry out a forensic triage of the asset to assess whether the system is compromised." Each CISA catalog entry states whether forensic triage applies. Patching alone does not answer whether an attacker got in first. Example: Adobe Commerce and Magento CVE-2026-75650 was exploited from 4 Sep 2026, three days before the 7 Sep advisory; the catalog entry of 8 Sep marks forensic triage Yes, due 11 Sep. (CVE-2026-75650 example from [CISA catalog entry](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-75650), read 19 Nov 2026) ([source](https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk), read 19 Nov 2026) ## Incident NIST CSRC defines an incident as "An occurrence that actually or potentially jeopardizes the confidentiality, integrity, or availability of an information system." A vulnerability alone is not an incident; exploitation, or reasonable suspicion of it, turns a patching task into incident response with a record and an owner. Example: For Cisco ISE CVE-2026-76460, disclosed 16 Sep 2026, Cisco advised that anyone who suspects compromise re-image the affected nodes, so a suspected hit became a rebuild. (example from [Cisco advisory cisco-sa-ISE-ABP-VNSW7Tn5](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/incident), read 19 Nov 2026) ## Indicator of compromise NIST SP 800-61r3, as shown in the NIST CSRC glossary, defines indicators of compromise as "Technical artifacts or observables that suggest that an attack is imminent or is currently underway or that a compromise may have already occurred." Vendors publish them in advisories, and they go stale as attackers change infrastructure. Example: For CVE-2026-0322, disclosed 14 Oct 2026, Palo Alto Networks tied three addresses to limited exploit attempts: 66[.]42[.]72[.]14, 208[.]83[.]239[.]114 and 149[.]248[.]47[.]234. (example from [Palo Alto Networks advisory CVE-2026-0322](https://security.paloaltonetworks.com/CVE-2026-0322), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/indicator_of_compromise), read 19 Nov 2026) ## Known Exploited Vulnerabilities catalog CISA states it "maintains the authoritative source of vulnerabilities that have been exploited in the wild." Each entry carries a date added. The Disclosure Clock treats that date as an upper bound on first exploitation when no earlier dated report exists, since exploitation can start before CISA lists the flaw. Example: VMware ESX CVE-2026-41709 was disclosed on 29 Jul 2026 and added to the catalog on 28 Oct 2026, 91 days later, so 28 Oct is an upper bound on when exploitation began. (example from [CISA catalog entry](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-41709), read 19 Nov 2026) ([source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog), read 19 Nov 2026) ## Mitigation NIST SP 800-216, in the NIST CSRC glossary, defines mitigation as "The temporary reduction or lessening of the impact of a vulnerability or the likelihood of its exploitation." A mitigation buys time for the patch. It needs an owner, a stated cost and a date on which it is removed. Example: For ScreenConnect CVE-2026-84869, disclosed 3 Sep 2026, ConnectWise advised deselecting the TransferFiles permission in every role until 26.6.5, released 8 Sep 2026; the cost is that technicians cannot transfer files in sessions. (example from [ConnectWise bulletin of 8 Sep 2026](https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/mitigation), read 19 Nov 2026) ## Patch management NIST SP 800-40 Rev. 4 opens its abstract: "Enterprise patch management is the process of identifying, prioritizing, acquiring, installing, and verifying the installation of patches, updates, and upgrades throughout an organization." Emergency patching runs the same steps on a shorter clock, and verification is the step teams most often skip. Example: Microsoft fixed [Windows](/playbooks/windows-endpoint-os/) AFD CVE-2026-68820 in the cumulative updates of 11 Aug 2026, the day the flaw was disclosed and exploited, so installation and verification had to start on day zero. (example from [Microsoft August 2026 CVRF feed](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug), read 19 Nov 2026) ([source](https://csrc.nist.gov/pubs/sp/800/40/r4/final), read 19 Nov 2026) ## Remediation BOD 26-04 treats an asset as remediated when "the vulnerability is eliminated through patching, decommissioning the system, or another action." A mitigation that lowers risk without removing the flaw does not meet that test, although the directive lets removing a system from the internet change the timeline that applies. Example: FortiMail CVE-2026-104286, disclosed 1 Oct 2026, is remediated by upgrading to 8.0.2, 7.6.7 or 7.4.9 or later; turning off the IBE service only mitigates it. (example from [Fortinet advisory FG-IR-26-175](https://fortiguard.fortinet.com/psirt/FG-IR-26-175), read 19 Nov 2026) ([source](https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk), read 19 Nov 2026) ## Security advisory The OASIS Common Security Advisory Framework 2.0 describes security advisories as "structured information on products, vulnerabilities and the status of impact and remediation among interested parties." Vendor advisories are the source for affected and fixed versions. Read the advisory itself before any summary, because vendors revise advisories after first publication. Example: The Palo Alto Networks advisory for CVE-2026-0322, published 14 Oct 2026, lists fixed builds per PAN-OS train and states that no workaround exists. (example from [Palo Alto Networks advisory CVE-2026-0322](https://security.paloaltonetworks.com/CVE-2026-0322), read 19 Nov 2026) ([source](https://docs.oasis-open.org/csaf/csaf/v2.0/os/csaf-v2.0-os.html), read 19 Nov 2026) ## Software bill of materials NIST CSRC defines a software bill of materials as "A formal record containing the details and supply chain relationships of various components used in building software." In a disclosure it answers which applications embed the affected library, and its date shows how current that answer is. Example: The Node.js advisory of 21 Oct 2026 for CVE-2026-111497 covers any [software embedding Brotli](/playbooks/dependency/) 1.0.3 to 1.2.0; fixes are Brotli 1.0.10, 1.1.1 and 1.2.1, and a Node.js upgrade fixes only the bundled copy. (example from [Node.js October 2026 security releases](https://nodejs.org/en/blog/vulnerability/october-2026-security-releases), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/software_bill_of_materials), read 19 Nov 2026) ## Zero-day vulnerability NIST CSRC defines a zero-day attack, not the vulnerability itself, as "An attack that exploits a previously unknown hardware, firmware, or software vulnerability." A zero-day vulnerability is the flaw such an attack uses. On the Disclosure Clock the practical test is dated exploitation before the vendor advisory, which leaves defenders no patch on the day exploitation starts. Example: BeyondTrust CVE-2026-102764 was exploited from 29 Aug 2026, 47 days before the advisory of 15 Oct 2026; Adobe CVE-2026-75650 was exploited from 4 Sep 2026 and had its advisory on 7 Sep. (examples from [BeyondTrust advisory BT26-05](https://www.beyondtrust.com/trust-center/security-advisories/bt26-05) and [Fastly on Style Smuggler](https://www.fastly.com/blog/what-is-style-smuggler-cve-2026-75650), read 19 Nov 2026) ([source](https://csrc.nist.gov/glossary/term/zero_day_attack), read 19 Nov 2026) ## Related pages - [The Disclosure Clock: median 2 days to exploitation](/clock/) ## Sources - https://csrc.nist.gov/glossary/term/attack_surface (read 19 Nov 2026) - https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/compensating_security_control (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/common_vulnerabilities_and_exposures (read 19 Nov 2026) - https://www.first.org/cvss/v4-0/specification-document (read 19 Nov 2026) - https://cwe.mitre.org/about/faq.html (read 19 Nov 2026) - https://zerodayplaybook.com/#zp-clock (read 19 Nov 2026) - https://www.cisa.gov/news-events/directives/v1-ed-26-03-mitigate-vulnerabilities-cisco-sd-wan-systems (read 19 Nov 2026) - https://www.first.org/epss/ (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/incident (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/indicator_of_compromise (read 19 Nov 2026) - https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/mitigation (read 19 Nov 2026) - https://csrc.nist.gov/pubs/sp/800/40/r4/final (read 19 Nov 2026) - https://docs.oasis-open.org/csaf/csaf/v2.0/os/csaf-v2.0-os.html (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/software_bill_of_materials (read 19 Nov 2026) - https://csrc.nist.gov/glossary/term/zero_day_attack (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/glossary/ --- # About Zero Day Playbook About, updated 19 Nov 2026 **The short answer.** Zero Day Playbook is a reference for the first 72 hours after a critical disclosure. The edition of 19 Nov 2026 holds a 72-hour checklist, 13 playbooks, the Disclosure Clock of 33 bugs from 2026 with a dated first exploitation and 20 [glossary](/glossary/) terms. Two readers use it. One is the person on call when an advisory lands, who has to decide before the next shift. The other is the security lead who briefs the board by noon and needs a number, an owner and a date. The playbooks work as runbooks. Each step has an owner, a window (0 to 4, 4 to 24 or 24 to 72 hours) and a source, so a team can copy a playbook into its own procedure and change the names. Thirteen playbook pages are written, two by role (communications, executive) and eleven by asset type, from endpoint fleets to email gateways. The checklist on the home page filters by the role you pick (communications or executive): it lists that role's steps, then the steps of the asset you pick, marks which are the role's own and shows the rest as hand-offs to other owners such as the security lead and platform engineering. It puts the steps in order and follows the hours since the advisory. The Disclosure Clock is the site's own dataset. It lists 33 bugs of 2026 with a dated first exploitation. The CSV and JSON hold those 33 rows, each with the disclosure date, the date CISA added the bug to its Known Exploited Vulnerabilities catalog, the first dated exploitation and the days between, with a median of 2 days. Headline bugs of 2026 with no dated first exploitation have no row. The downloads are [clock.csv](/datasets/clock.csv) and [clock.json](/datasets/clock.json), also linked in the Disclosure Clock section of the [home page](/). Sources are primary where possible: vendor advisories, CISA directives and its catalog, the National Vulnerability Database, and NIST SP 800-40 Rev. 4 and SP 800-61 Rev. 3. The [editorial policy](/editorial-policy/) sets out how dates are taken, what the labels mean and what is never published. The first edition appeared on 10 Nov 2026. The edition of 19 Nov 2026 re-read the Clock against NVD, the CISA feed and vendor advisories, removed three entries, merged overlapping pages and rewrote the policy pages. The editorial team writes and checks every page and answers at editor@zerodayplaybook.com; see [contact](/contact/). ## Related pages - [Editorial policy](/editorial-policy/) - [Contact](/contact/) - [Privacy](/privacy/) - [Terms](/terms/) ## Sources - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - NIST SP 800-40 Rev. 4: https://csrc.nist.gov/pubs/sp/800/40/r4/final (read 19 Nov 2026) - NIST SP 800-61 Rev. 3: https://csrc.nist.gov/pubs/sp/800/61/r3/final (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/about/ --- # How Zero Day Playbook is written Editorial policy, updated 19 Nov 2026 **The short answer.** Zero Day Playbook is a reference that lists and does not rank: bugs are ordered by date or by days to exploitation. One editorial team, reached at editor@zerodayplaybook.com, writes and checks the edition of 19 Nov 2026. Each page states the question it answers and sources every date, version, rating and count on it. Sources. Dated claims come from primary pages: vendor advisories and release notes, the CISA directive BOD 26-04 and Known Exploited Vulnerabilities catalog, NVD records, and NIST SP 800-40 Rev. 4 and SP 800-61 Rev. 3. Each is linked with the date it was opened. Quotations are exact and short, and each names its author. A named security firm is used for an exploitation date when it reported first. Clock dates. Disclosure is the date on the vendor advisory, or the NVD published date when the vendor states none. First exploitation is the earliest dated public report from the vendor or a named security firm. With no such report it is the date of exploitation marked active in CISA's SSVC data in the NVD record, when that data is present. With neither it is the date CISA added the bug to its catalog, which is an upper bound on the first exploitation date, because exploitation can start before CISA lists a bug. In the edition of 19 Nov 2026, 16 rows have a first exploitation date equal to the catalog date and 17 rest on a report dated before the catalog date, 12 from a vendor or named security firm and 5 from CISA SSVC data in NVD records; the source column of the CSV names the page for each. A negative number means exploitation was reported before disclosure, and a bug with no dated exploitation gets no row. Four of the 37 headline bugs have none on 19 Nov 2026: the fastjson 1.x flaw CVE-2026-16723 (disclosed 21 Jul), the Next.js flaw CVE-2026-75604 (25 Aug), the Linux kernel net/unix flaw CVE-2026-80521 (26 Aug) and the Satori and Next.js flaw CVE-2026-94545 (30 Sep). A blank date does not show that a bug is unexploited. Labels. A fact a vendor states about itself is labeled the vendor's claim, and its words are quoted. A fact from anyone else is labeled a third-party source. Where a page says nothing on a point, the entry reads none found with the read date; that describes the pages read, not the world. Order. The Clock sorts by measured days, ties by disclosure date. The playbook list on the home page puts the two role playbooks first, then the eleven asset types, in the order the page shows rather than A to Z. The [glossary](/glossary/) runs A to Z. No page carries a rating, a weight or a position number for any vendor or bug. Read dates. Every source link carries the date it was read. The playbooks are re-read against the guidance they cite at least every 90 days, and the Clock gains a row once a first exploitation date is published. The Clock and its CVE records were re-read on 19 Nov 2026 against NVD, the CISA feed and vendor advisories. Never published. The site publishes no exploit code, no proof-of-concept steps and no scanning instructions. Text stays defensive and gives preconditions, detection, mitigation and fix only. It publishes no rating of a vendor and no date it could not source. Changes. A Clock date moves only when an earlier public report appears. A reader who finds a source that no longer says what a page says can send the address, and the page is checked and changed in the open. ## Related pages - [About](/about/) - [Contact](/contact/) - [Privacy](/privacy/) - [Terms](/terms/) ## Sources - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - CISA BOD 26-04, Prioritizing Security Updates Based on Risk: https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk (read 19 Nov 2026) - NIST SP 800-40 Rev. 4: https://csrc.nist.gov/pubs/sp/800/40/r4/final (read 19 Nov 2026) - NIST SP 800-61 Rev. 3: https://csrc.nist.gov/pubs/sp/800/61/r3/final (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/editorial-policy/ --- # Contact Zero Day Playbook Contact, updated 19 Nov 2026 **The short answer.** Write to [editor@zerodayplaybook.com](mailto:editor@zerodayplaybook.com) to report an error, send an earlier exploitation report or ask to reuse the data. Email is the only way to reach the editorial team: the site has no forms, no accounts and no newsletter list. Expect a reply within three working days. To report a wrong date, version or claim, send the page path, the claim as printed, what you think it should say and a public URL that supports it: a vendor advisory, a CISA entry or an NVD record. A confirmed correction is made on the page. For a question about the Disclosure Clock, name the CVE and the column. To report exploitation earlier than the Clock shows, send the CVE, the date of the first public report you found and its link. If that date is earlier, the row changes and the old date stays in the log. News tips should link a primary page: a vendor advisory, a release note or a catalog entry. A vendor that wants a fact about its own tool changed or added should send a public URL that states it. The fact is added with that URL and a read date, labeled as the vendor's claim. The Clock is published as CSV and JSON under CC BY 4.0 with its cite line. Ask by email if you need another cut of the rows. Do not send exploit code or proof-of-concept material. Report vulnerabilities to the vendor named on the relevant advisory. ## Related pages - [About](/about/) - [Editorial policy](/editorial-policy/) - [Privacy](/privacy/) - [Terms](/terms/) ## Sources - CISA Known Exploited Vulnerabilities catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog (read 19 Nov 2026) - NVD CVE API: https://nvd.nist.gov/developers/vulnerabilities (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/contact/ --- # Privacy Privacy, updated 19 Nov 2026 **The short answer.** Zero Day Playbook sets no cookies, runs no analytics and loads no third-party scripts. It has no forms, no accounts and no newsletter list. Fonts are self-hosted, so no font is requested from another host. The only record of a visit is the access log of the network that delivers the pages, kept for 400 days. The site is a set of static HTML files on Amazon S3, served by Amazon CloudFront. The checklist and any other tool on the pages run in your browser, and the values you enter are not sent anywhere. CloudFront writes each request to a private S3 bucket in AWS's US East (N. Virginia) region. A log line records when the request came, which edge served it, the visitor's IP address, the path asked for, the response code, the referring page and the browser string; the [AWS standard logging reference](https://docs.aws.amazon.com/AmazonCloudFront/latest/DeveloperGuide/standard-logs-reference.html) lists every field (read 19 Nov 2026). The logs are used for two counts: page views per page, and crawler visits, separated by user agent where it allows. They are totals, not trails of readers. No profile of any reader is built, and the log carries no account or identity, although an IP address in it can still point to a person or an office. For that reason the bucket is private, and a lifecycle rule set on 19 Nov 2026 deletes the log files after 400 days. Email sent to editor@zerodayplaybook.com is kept to answer it and to make any correction it causes, and is deleted on request. To ask about the log lines tied to your address, or to have an email deleted, write to the same address. ## Related pages - [About](/about/) - [Editorial policy](/editorial-policy/) - [Contact](/contact/) - [Terms](/terms/) ## Sources - AWS CloudFront standard logging reference: https://docs.aws.amazon.com/AmazonCloudFront/latest/DeveloperGuide/standard-logs-reference.html (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/privacy/ --- # Terms of use Terms, updated 19 Nov 2026 **The short answer.** The Disclosure Clock data, as [CSV](/datasets/clock.csv) and [JSON](/datasets/clock.json), may be reused under CC BY 4.0 with the cite line below. The text may be quoted with a citation and a link. Pages are a reference, not advice for a particular system. The cite line is: Zero Day Playbook, The Disclosure Clock, 33 dated rows of the 37 headline bugs of 2026, edition 19 Nov 2026, zerodayplaybook.com/. The other 4 headline bugs, named on the [editorial policy](/editorial-policy/), have no dated exploitation on that date and so no row. Keep the edition date, since rows move when an earlier report appears. Information only. The site gives general information, not legal or incident response advice. Advisories are revised after publication, so before you act on a version number or a date, open the vendor's own page and the fact's source, and do the same before relying on it for a purchase, a patch decision or a compliance filing. If the source and the page differ, write to editor@zerodayplaybook.com and the page will be corrected. The playbooks are starting points for a runbook. Adapt each step to your own systems, contracts and regulators. A playbook line names an owner and a window as an example, and your organization sets its own. The Clock measures public reports. A row with a dated exploitation report shows when someone first said so in public, and exploitation may have begun earlier. A Clock median is not a forecast for the next bug. Third-party material. Vendor facts, quotations from advisories and release notes, and product and company names belong to their owners. They appear as short extracts with a link, so a reader can tell which product a fact is about. Linked sites are theirs and can change after the date they were read. Liability. The site and its data are provided as they stand on the date of the edition, without warranty of any kind. To the extent the law allows, the publisher is not liable for loss that comes from relying on them. ## Related pages - [About](/about/) - [Editorial policy](/editorial-policy/) - [Contact](/contact/) - [Privacy](/privacy/) ## Sources - Creative Commons CC BY 4.0: https://creativecommons.org/licenses/by/4.0/ (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/terms/ --- # Median 2 days from disclosure to exploitation across 33 headline bugs Disclosure Clock, read 19 Nov 2026 **The short answer.** The Disclosure Clock lists 37 headline bugs of 2026. Thirty-three have a dated first exploitation, 12 of them on or before disclosure day, and the median gap is 2 days. Dates come from vendor advisories, the CISA catalog and named reports, re-read on 19 Nov 2026. The [response guide](/guides/what-to-do-when-a-zero-day-is-disclosed/) covers the first hours. | CVE | Days | Product | Disclosed | First exploited | Added to KEV | Basis of the date | | --- | --- | --- | --- | --- | --- | --- | | CVE-2026-102764 | -47 (exploited 47 days before disclosure) | BeyondTrust Remote Support | 15 Oct 2026 | [29 Aug 2026](https://www.beyondtrust.com/blog/entry/bt26-05) | 16 Oct 2026 | named report | | CVE-2026-75650 | -3 (exploited 3 days before disclosure) | Adobe Commerce and Magento Open Source | 7 Sep 2026 | [4 Sep 2026](https://sansec.io/research/stylesmuggler-0day) | 8 Sep 2026 | named report | | CVE-2026-68820 | 0 | Microsoft Windows AFD for WinSock | 11 Aug 2026 | [11 Aug 2026](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug) | 11 Aug 2026 | named report | | CVE-2026-87491 | 0 | Google Chromium V8 | 4 Sep 2026 | [4 Sep 2026](https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security) | 9 Sep 2026 | named report | | CVE-2026-81963 | 0 | Microsoft Windows Update Stack | 8 Sep 2026 | [8 Sep 2026](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep) | 8 Sep 2026 | named report | | CVE-2026-85880 | 0 | Microsoft Windows ALPC | 8 Sep 2026 | [8 Sep 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-85880) | 8 Sep 2026 | CISA date (KEV) | | CVE-2026-76460 | 0 | Cisco Identity Services Engine | 16 Sep 2026 | [16 Sep 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | 16 Sep 2026 | CISA date (KEV) | | CVE-2026-94127 | 0 | F5 BIG-IP APM | 22 Sep 2026 | [22 Sep 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-94127) | 22 Sep 2026 | CISA date (KEV) | | CVE-2026-104286 | 0 | Fortinet FortiMail | 1 Oct 2026 | [1 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286) | 1 Oct 2026 | CISA date (KEV) | | CVE-2026-0322 | 0 | Palo Alto Networks PAN-OS | 14 Oct 2026 | [14 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322) | 14 Oct 2026 | CISA date (KEV) | | CVE-2026-109618 | 0 | PHP Group PHP | 21 Oct 2026 | [21 Oct 2026](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | 26 Oct 2026 | named report | | CVE-2026-76418 | 0 | Cisco Secure Firewall ASA and FTD | 28 Oct 2026 | [28 Oct 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76418) | 29 Oct 2026 | CISA date (SSVC) | | CVE-2026-46300 | 1 | Linux Kernel | 13 May 2026 | [14 May 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-46300) | 28 Oct 2026 | CISA date (SSVC) | | CVE-2026-88779 | 1 | Citrix NetScaler ADC and Gateway | 3 Oct 2026 | [4 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-88779) | 4 Oct 2026 | CISA date (KEV) | | CVE-2026-97229 | 1 | GitLab CE/EE | 12 Oct 2026 | [13 Oct 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-97229) | 29 Oct 2026 | CISA date (SSVC) | | CVE-2026-19868 | 1 | SonicWall SonicOS SSLVPN | 5 Nov 2026 | [6 Nov 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-19868) | 6 Nov 2026 | CISA date (SSVC) | | CVE-2026-64849 | 2 | MLflow Tracking Server | 17 Aug 2026 | [19 Aug 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-64849) | 19 Aug 2026 | CISA date (KEV) | | CVE-2026-123158 | 2 | OpenBSD OpenSSH | 5 Nov 2026 | [7 Nov 2026](https://sshole.events.greynoise.io/) | 9 Nov 2026 | named report | | CVE-2026-42018 | 3 | JFrog Artifactory | 12 Aug 2026 | [15 Aug 2026](https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201) | 11 Sep 2026 | named report | | CVE-2026-82329 | 4 | JFrog Artifactory | 28 Aug 2026 | [1 Sep 2026](https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity) | 2 Sep 2026 | named report | | CVE-2026-109619 | 5 | PHP Group PHP | 21 Oct 2026 | [26 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-109619) | 26 Oct 2026 | CISA date (KEV) | | CVE-2026-111497 | 5 | Node.js Brotli decoder | 21 Oct 2026 | [26 Oct 2026](https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/) | 27 Oct 2026 | named report | | CVE-2026-119977 | 6 | WordPress Core | 28 Oct 2026 | [3 Nov 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-119977) | 3 Nov 2026 | CISA date (KEV) | | CVE-2026-84869 | 7 | ConnectWise ScreenConnect | 3 Sep 2026 | [10 Sep 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-84869) | 11 Sep 2026 | CISA date (SSVC) | | CVE-2026-21589 | 10 | Atlassian Data Center products | 5 Oct 2026 | [15 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-21589) | 15 Oct 2026 | CISA date (KEV) | | CVE-2026-34486 | 15 | Apache Software Foundation Apache Tomcat | 9 Apr 2026 | [24 Apr 2026](https://socradar.io/blog/snowlight-government-chinese-campaign/) | 4 Aug 2026 | named report | | CVE-2026-9198 | 15 | IBM Langflow OSS | 2 Jul 2026 | [17 Jul 2026](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-9198) | 4 Aug 2026 | CISA date (SSVC) | | CVE-2026-42016 | 19 | JFrog Artifactory | 27 Jul 2026 | [15 Aug 2026](https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201) | 11 Sep 2026 | named report | | CVE-2026-19490 | 21 | Citrix NetScaler ADC and Gateway | 19 Aug 2026 | [9 Sep 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19490) | 9 Sep 2026 | CISA date (KEV) | | CVE-2026-65660 | 44 | Microsoft SharePoint Server | 11 Aug 2026 | [24 Sep 2026](https://blog.previdian.com/cve-2026-65660-previdian-observes-two-stage-sharepoint-exploitation-attempts/) | 25 Sep 2026 | named report | | CVE-2026-59822 | 50 | BerriAI LiteLLM | 8 Jul 2026 | [27 Aug 2026](https://www.wiz.io/blog/ai-infrastructure-honeypot) | 2 Sep 2026 | named report | | CVE-2026-41709 | 91 | Broadcom VMware ESX | 29 Jul 2026 | [28 Oct 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-41709) | 28 Oct 2026 | CISA date (KEV) | | CVE-2026-53266 | 109 | Linux kernel | 1 Jun 2026 | [18 Sep 2026](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-53266) | 18 Sep 2026 | CISA date (KEV) | Ordered by days, fewest first, then by disclosure date. Median 2 days. First exploitation is the earliest dated public report; the CISA catalog date is the latest it can be. ## What the days show The longest gap is 109 days, for the Linux flaw CVE-2026-53266, disclosed 1 Jun and listed by CISA on 18 Sep 2026. Two of the four rows over 30 days rest on the catalog date, which is an upper bound on the first exploitation date, so their true gaps may be shorter. Two rows are negative: CVE-2026-102764 and CVE-2026-75650. CVE-2026-87491 sits at day 0: Microsoft's Edge release note of 4 Sep 2026 is the first vendor advisory to name the CVE, and it says the Chromium team reported an exploit in the wild, so disclosure and first exploitation share that date. Google's [Chrome note](https://chromereleases.googleblog.com/2026/09/stable-channel-update-for-desktop_0808145027.html) followed on 8 Sep 2026. For the first 72 hours the figures say three things. By the end of day 3, 19 of the 33 bugs, more than half, were exploited, so plan in days, not weeks. Sixteen were exploited within one day and 24 within a week, so a first control on every exposed asset inside one day and a patch inside a week cover most observed attacks (see [mitigations that buy time](/guides/mitigations-that-buy-time/) and the 18 playbooks). The tools page lists what 14 vendors show for a new disclosure. And 9 bugs were first exploited 8 or more days after disclosure, so an item closed as not exploited should reopen when the CISA catalog lists it. The table covers headline bugs that drew an advisory and a catalog entry, not all vulnerabilities, and it is no forecast for any one CVE. Dates follow fixed rules. Disclosure is the date of the earliest vendor advisory that names the bug, or the NVD published date when no vendor advisory does. First exploitation is the earliest dated public report from the vendor or a named security firm; with none, it is the date of CISA's exploitation stamp in the NVD record where one exists (basis "CISA date (SSVC)", 6 rows: CVE-2026-46300, CVE-2026-97229, CVE-2026-84869, CVE-2026-9198, CVE-2026-19868 and CVE-2026-76418), and otherwise the date the bug was added to the catalog (basis "CISA date (KEV)", 13 rows, including CVE-2026-85880). In this edition 19 rows rest on CISA data and 14 on a named report. CVE-2026-42016 shares Wiz's Artifactory report with CVE-2026-42018, so the source list names it once. Days are calendar days, so a bug disclosed at 23:00 and exploited at 01:00 counts as one day. CVE-2026-82329, the Artifactory bypass, was disclosed 28 Aug 2026; Fastly saw no exploitation that day, scanning from offensive-security services on 31 Aug and mass exploitation from 1 Sep, so the gap is 4 days. BeyondTrust dated exploitation of CVE-2026-102764 to 29 Aug 2026, 47 days before its advisory of 15 Oct, which only a hunt after the advisory can find. Four bugs have no dated first exploitation on 19 Nov 2026 and so no row: the fastjson 1.x flaw CVE-2026-16723, the Next.js flaw CVE-2026-75604, the Linux AF_UNIX flaw CVE-2026-80521 and the Satori flaw CVE-2026-94545. A blank date does not show that a bug is unexploited. ## Download the data Download the 33 dated rows as [CSV](/datasets/clock.csv) or [JSON](/datasets/clock.json) under CC BY 4.0. One row per dated bug, with the columns cve, vendor, product, disclosed, first_exploited, kev_added, days (negative before disclosure) and source (the page that dates the first exploitation). Cite them as: Zero Day Playbook, The Disclosure Clock, 33 dated rows of the 37 headline bugs of 2026, edition 19 Nov 2026, zerodayplaybook.com/clock/. - [Disclosure Clock, CSV](/datasets/clock.csv) - [Disclosure Clock, JSON](/datasets/clock.json) ## Related pages - [Glossary](/glossary/) ## Sources - CVE-2026-102764: first exploitation report: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - CVE-2026-102764: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-102764 (read 19 Nov 2026) - CVE-2026-75650: first exploitation report: https://sansec.io/research/stylesmuggler-0day (read 19 Nov 2026) - CVE-2026-75650: NVD record carrying Adobe's advisory data for APSB26-146 (affected versions, hotfix, CVSS): https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-75650 (read 19 Nov 2026) - CVE-2026-68820: first exploitation report: https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug (read 19 Nov 2026) - CVE-2026-68820: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-68820 (read 19 Nov 2026) - CVE-2026-87491: first exploitation report: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security (read 19 Nov 2026) - CVE-2026-87491: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-87491 (read 19 Nov 2026) - CVE-2026-81963: first exploitation report: https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep (read 19 Nov 2026) - CVE-2026-81963: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-81963 (read 19 Nov 2026) - CVE-2026-85880: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-85880 (read 19 Nov 2026) - CVE-2026-85880: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-85880 (read 19 Nov 2026) - CVE-2026-76460: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460 (read 19 Nov 2026) - CVE-2026-76460: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76460 (read 19 Nov 2026) - CVE-2026-94127: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-94127 (read 19 Nov 2026) - CVE-2026-94127: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-94127 (read 19 Nov 2026) - CVE-2026-104286: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286 (read 19 Nov 2026) - CVE-2026-104286: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-104286 (read 19 Nov 2026) - CVE-2026-0322: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322 (read 19 Nov 2026) - CVE-2026-0322: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-0322 (read 19 Nov 2026) - CVE-2026-109618: first exploitation report: https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell (read 19 Nov 2026) - CVE-2026-109618: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-109618 (read 19 Nov 2026) - CVE-2026-76418: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76418 (read 19 Nov 2026) - CVE-2026-46300: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-46300 (read 19 Nov 2026) - CVE-2026-88779: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-88779 (read 19 Nov 2026) - CVE-2026-88779: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-88779 (read 19 Nov 2026) - CVE-2026-97229: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-97229 (read 19 Nov 2026) - CVE-2026-19868: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-19868 (read 19 Nov 2026) - CVE-2026-64849: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-64849 (read 19 Nov 2026) - CVE-2026-64849: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-64849 (read 19 Nov 2026) - CVE-2026-123158: first exploitation report: https://sshole.events.greynoise.io/ (read 19 Nov 2026) - CVE-2026-123158: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026) - CVE-2026-42018: first exploitation report: https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201 (read 19 Nov 2026) - CVE-2026-42018: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-42018 (read 19 Nov 2026) - CVE-2026-82329: first exploitation report: https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity (read 19 Nov 2026) - CVE-2026-82329: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-82329 (read 19 Nov 2026) - CVE-2026-109619: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-109619 (read 19 Nov 2026) - CVE-2026-109619: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-109619 (read 19 Nov 2026) - CVE-2026-111497: first exploitation report: https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/ (read 19 Nov 2026) - CVE-2026-111497: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-111497 (read 19 Nov 2026) - CVE-2026-119977: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-119977 (read 19 Nov 2026) - CVE-2026-119977: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-119977 (read 19 Nov 2026) - CVE-2026-84869: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-84869 (read 19 Nov 2026) - CVE-2026-21589: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-21589 (read 19 Nov 2026) - CVE-2026-21589: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-21589 (read 19 Nov 2026) - CVE-2026-34486: first exploitation report: https://socradar.io/blog/snowlight-government-chinese-campaign/ (read 19 Nov 2026) - CVE-2026-34486: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-34486 (read 19 Nov 2026) - CVE-2026-9198: first exploitation report: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-9198 (read 19 Nov 2026) - CVE-2026-42016: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-42016 (read 19 Nov 2026) - CVE-2026-19490: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19490 (read 19 Nov 2026) - CVE-2026-19490: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-19490 (read 19 Nov 2026) - CVE-2026-65660: first exploitation report: https://blog.previdian.com/cve-2026-65660-previdian-observes-two-stage-sharepoint-exploitation-attempts/ (read 19 Nov 2026) - CVE-2026-65660: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-65660 (read 19 Nov 2026) - CVE-2026-59822: first exploitation report: https://www.wiz.io/blog/ai-infrastructure-honeypot (read 19 Nov 2026) - CVE-2026-59822: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-59822 (read 19 Nov 2026) - CVE-2026-41709: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-41709 (read 19 Nov 2026) - CVE-2026-41709: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-41709 (read 19 Nov 2026) - CVE-2026-53266: first exploitation report: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-53266 (read 19 Nov 2026) - CVE-2026-53266: NVD record: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-53266 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/clock/ --- # The communications playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Communications writes only what the incident record supports. The page gives a complete holding statement, a customer update, a support script and the legal clocks quoted from their texts: SEC Form 8-K Item 1.05 four business days, NIS2 24-hour early warning, GDPR 72 hours. The worked case is BeyondTrust, exploited 47 days before disclosure. Communications is the one role that must publish before the facts are in. The security lead may need days to learn whether anyone got in, while customers, staff and sometimes regulators need words in hours. The risk runs both ways. A statement that says "no impact" on hour three is wrong if an intruder arrived first, and silence leaves the story to others. The fix is to write only claims that carry a source and a time, and to say what is unknown. Zero-days make this harder in one way. The disclosure date is not the exposure date. BeyondTrust and Mandiant dated the first exploitation of CVE-2026-102764 to 29 Aug 2026, 47 days before the advisory of 15 Oct 2026, so a customer's "we found nothing" is only as good as the logs behind it and the date they start from. Separately, legal clocks run from the moment you become aware of an incident or determine one is material, never from the date of an advisory, so the person who writes the statement must also see the clock table. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Communications | Get the facts sheet from the security lead and mark each line Confirmed, Unconfirmed or Unknown, with the source and the time. Add a basis to every "no evidence" claim: which logs, which systems, from which date to which date. A claim that rests on logs stored only on the affected device is weak, because actors known for appliance zero-days delete local logs.
Every sentence you publish later is checked against this sheet, and a missing basis is what turns a true statement into a misleading one. | The facts sheet with version numbers, and the basis line for each no-evidence claim. | | 0 to 1 hour | Communications | Name one spokesperson and one approver, plus a legal reviewer who can answer within the hour. Route every press, customer and regulator inquiry to the spokesperson. Staff and partners get no other channel, and no one posts on social media about the incident.
Two people saying different things in the first hours cannot be taken back, and a named reviewer is what keeps drafting inside the 4-hour window. | Names, mobile numbers, the approval route and the time each was confirmed. | | 1 to 4 hours | Communications | Write the holding statement from this template and publish it only when the approver signs.
[Date, time, time zone] Update on [CVE ID], a vulnerability in [product and versions].
What we know: On [date], [vendor or source] disclosed [CVE ID], which lets an attacker [effect in one plain clause]. [The vendor or CISA] reports it is being exploited.
What we are doing: We use [product] in [named systems]. We are [applying the vendor's mitigation or fix] and reviewing [logs] from [lookback start date] for signs of misuse.
What we have found so far: [Option A: We have found no evidence of misuse in [logs, systems] from [date] to [time]. The review is not finished. / Option B: We have found activity consistent with exploitation on [systems] and have started our incident response.]
What you can do: [one action for customers, such as check the version or rotate credentials].
Next update: [time, time zone] at [URL], or sooner if this changes. Contact: [name or address].
Leave out "no impact", "fully secure" and "isolated incident" until the triage report supports them.
The statement exists to give the first answer and the time of the second, and the template forces the unknowns onto the page. | The published text, the time, the approver, and the facts-sheet version each line came from. | | 1 to 4 hours | Communications | Brief support and account teams with a three-sentence script and an escalation list. Sentence 1: "[Vendor] disclosed a vulnerability in [product] on [date], and it is being exploited." Sentence 2: "We are [applying the fix or mitigation] and checking for misuse, and we will update at [time]." Sentence 3: "If you see [specific sign], tell us at [address]." Questions on data, impact, liability, causes or dates go to the spokesperson, and the person answering writes the question down.
A customer who asks three people should hear one answer, and the written questions show what the next update must cover. | The script version, the list of staff briefed, and the questions logged with time. | | 1 to 4 hours | Communications | Put the notification clocks in the incident record with counsel, one line per regime, each with a trigger, a deadline in UTC and an owner.
SEC Form 8-K Item 1.05: "within four business days after the registrant determines that it has experienced a material cybersecurity incident"; the materiality determination "must be made without unreasonable delay after discovery of the incident"; missing information goes in an amendment within four business days of becoming available.
NIS2 Article 23(4), essential and important entities: an early warning "within 24 hours of becoming aware of the significant incident"; an incident notification "within 72 hours"; a final report "not later than one month after the submission of the incident notification."
GDPR Article 33(1): to the supervisory authority "without undue delay and, where feasible, not later than 72 hours after having become aware of it," unless the breach "is unlikely to result in a risk," and a later filing "shall be accompanied by reasons for the delay"; Article 33(2): a processor tells the controller "without undue delay"; Article 34(1): data subjects are told "without undue delay" when the risk is high.
Contracts: list the hour-based notice clauses in your customer agreements.
The shortest clock is the 24-hour early warning, so the table must exist before anyone knows whether it applies. | For each regime: applies yes or no, trigger event and time, deadline in UTC, filer, counsel's sign-off. | | 4 to 24 hours | Communications | Draft the regulator text now, so the filing is a copy and not a writing task. For NIS2 the early warning states, where applicable, "whether the significant incident is suspected of being caused by unlawful or malicious acts or could have a cross-border impact." For GDPR, Article 33(3) requires the nature of the breach with the categories and approximate number of data subjects and records, the contact point, the likely consequences and the measures taken or proposed. Article 33(4) allows the information "in phases without undue further delay."
The filing deadline falls inside the next 24 to 72 hours, and a draft with gaps marked Unknown can go out on time while the facts arrive. | The draft, each gap marked Unknown, the reviewer, and the time the filing was sent with the reference number. | | 4 to 24 hours | Communications | Publish update 2 to customers at the time promised in the holding statement, even if nothing has changed, using this template.
Status: [Investigating / Mitigated / Fixed and verified].
Changed since the last update: [facts, dated].
Who is affected: [products, versions, customer groups], and who is not, with the basis.
What we found: [findings and the period reviewed, or "review not finished"].
What you should do: [steps with exact versions or settings].
Next update: [time, time zone].
An update on time with nothing new keeps customers from filling the silence, and keeping the same headings lets them see what changed. | The text, the time, the channels used and the number of customers who replied. | | 24 to 72 hours | Communications | Publish the fix-verified update only after the security lead supplies the verification output: state the date, the version or control now in place, and the lookback period searched. If any earlier statement turns out wrong, correct it in the same channel with the same prominence and say what changed. Do not use "out of an abundance of caution" to hide a fact that is known.
A fix claim is checked by customers who run the same product, and a correction made fast costs less than one found by someone else. | Verification evidence cited, the published text, the corrections issued and their times. | | 24 to 72 hours | Communications | Close the loop with employees and partners, then archive every statement with its time, approver and source line in a claims ledger of three columns: claim, evidence, basis. File the ledger with the incident record. Diary the NIS2 final report date, one month after the 72-hour notification, and any 8-K amendment date.
The ledger is what counsel and regulators ask for in a review, and the diary dates are the ones teams miss after the first 72 hours. | The claims ledger, the distribution list, and the diary dates for the final report and amendments. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet When the vendor has no fix, the statement cannot say "we have patched". Say what is in place (the mitigation, the date it went in, what it limits), what is not (the flaw remains), and when you will next check. Do not promise a fix date that the vendor has not given. If the mitigation limits access, name who loses it, because customers will find out within minutes. BeyondTrust did this in its own post: it reported what it had observed, what it withheld ("we are withholding the technical details") and what it could not prove, saying it lacked forensic proof of the AI-enabled tooling it suspected. If you take a service offline, publish what is unavailable, since when, and the time of your next review, not a promised time of return. A return time that slips costs more trust than the outage. Counsel should see the offline statement before it goes out, because a service outage can itself be a significant incident under NIS2 if it causes "severe operational disruption." ## CVE-2026-102764 worked through, hour by hour BeyondTrust and Mandiant disclosed on Thursday 15 Oct 2026 that a suspected China-nexus actor had exploited a flaw in Remote Support as a zero-day since 29 Aug 2026. The rows show what the vendor said and when, which is the standard a customer's own statements are compared with. | When | What happened | Source | | --- | --- | --- | | 29 Aug 2026 | The actor exploits the flaw as a zero-day against one BeyondTrust-hosted Remote Support instance and one self-hosted appliance. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 28 Sep 2026 | BeyondTrust's threat hunting finds the hosted instance; the company contains it "within 27 minutes" of first detection. Its post adds that its sensors fired but a third-party tool's automated triage "closed those alerts as false positives." | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 30 Sep 2026 | Mandiant shares findings from the self-hosted customer that confirm what BeyondTrust found on 28 Sep. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 8 Oct 2026 | BeyondTrust releases the BT26-05 patch. Hosted instances are patched automatically, and the advisory says all hosted customers were patched as of 9 Oct. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 15 Oct 2026 | Coordinated disclosure: advisory BT26-05, the CVE record, BeyondTrust's post and Mandiant's analysis are published the same day. BeyondTrust says it is "withholding the technical details" while appliances are patched. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 15 Oct 2026 (post text) | BeyondTrust: "we are only aware of the two impacted customers mentioned above, and have not seen evidence of adversary activity since August." The sentence is dated to the last update of the post. It also says the actor deletes logs and that off-box copies are often the only record. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 16 Oct 2026 | CISA adds the CVE to the KEV catalog with a due date of 19 Oct 2026 and forensic triage marked Yes. BeyondTrust updates its post at 12:00 EDT. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-102764) | | 4 Nov 2026 | BeyondTrust updates advisory BT26-05. | [source](https://www.beyondtrust.com/trust-center/security-advisories/bt26-05) | Three features of the vendor's text carry over to a customer's. Every claim is dated ("as of the time of the last update"), the limits of knowledge are stated, and the vendor admits its own miss. A customer on a self-hosted appliance would have two points of its own: the lookback starts on 29 Aug 2026, not 15 Oct, and any no-evidence line has to rest on logs held off the appliance. Now the clocks, on invented facts. Suppose a team confirms at 14:00 UTC on Thursday 15 Oct 2026 that an attacker took personal data from an appliance. The NIS2 early warning is due at 14:00 UTC on Friday 16 Oct and the incident notification at 14:00 UTC on Sunday 18 Oct, and the final report one month after that notification. The GDPR notice to the supervisory authority is also due by 14:00 UTC on Sunday 18 Oct, so the weekend is in the window. If the company determines on Thursday 15 Oct that the incident is material, the 8-K is due on Wednesday 21 Oct 2026, the fourth business day. The holding statement is due inside 4 hours, ahead of all of them. ## Related pages - [The executive playbook for the first 72 hours](/playbooks/executive/) - [Application runtime playbook: the first 72 hours when a flaw is in PHP or another language runtime on your servers](/playbooks/application-runtime/) - [The endpoint fleet playbook for the first 72 hours](/playbooks/endpoint-fleet/) ## Sources - BeyondTrust, ARMORY CASTLE and BT26-05: What Security Teams Need to Know: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - BeyondTrust Security Advisory BT26-05: https://www.beyondtrust.com/trust-center/security-advisories/bt26-05 (read 19 Nov 2026) - Mandiant and Google Threat Intelligence Group, ARMORY CASTLE Exploits BeyondTrust: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - CISA KEV entry for CVE-2026-102764: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-102764 (read 19 Nov 2026) - SEC final rule 33-11216, 88 FR 51896 (4 Aug 2023), Form 8-K Item 1.05: https://www.federalregister.gov/documents/full_text/html/2023/08/04/2023-16194.html (read 19 Nov 2026) - Directive (EU) 2022/2555 (NIS2), Article 23, Reporting obligations: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32022L2555 (read 19 Nov 2026) - Regulation (EU) 2016/679 (GDPR), Articles 33 and 34: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32016R0679 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/communications/ --- # The executive playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** The executive makes four calls in 72 hours: who decides, emergency change authority, downtime, and each risk acceptance. The page gives a filled five-line board brief, a risk acceptance record, the point where a securities filing clock starts, and the Cisco ISE advisory of 16 Sep 2026 worked hour by hour. An executive cannot patch anything, and the job in a zero-day is narrower than people expect: remove the delays that sit above the technical team. Those are approval to change production outside the calendar, a decision to give up revenue by taking a service down, and the signature on every risk that stays open. A weekly change board meets too late for most exploited bugs. The usual failure is waiting for certainty. For a bug on CISA's Known Exploited Vulnerabilities list the vendor has often already said that exploitation is under way, and the list entry itself starts the federal clock of three days for assets that need forensic triage. The facts the executive lacks in hour one are how many of your own systems are hit and whether anyone got in. Decide on what is known, name the unknowns in writing, and set a time to decide again. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Executive | Fix hour zero in writing: the earlier of the time the vendor published the advisory and the time your team learned of it, with its time zone. Name the one person who can approve downtime and the one who can accept risk, and set the bridge cadence: every 4 hours until the security lead reports the fix verified on every in-scope asset, then daily.
Every deadline in this table counts from hour zero, and without a named decision owner the first downtime question waits for a meeting. | Hour-zero time and basis, decision owner, risk owner, bridge schedule. | | 0 to 4 hours | Executive | Get the five-line brief from the security lead and use the same five lines for every later update. A filled example for the Palo Alto PAN-OS advisory of 14 Oct 2026, CVE-2026-0322 (the company figures are invented): Affected: 6 firewalls, 4 with the management interface reachable from the internet. Exploitation: confirmed by Palo Alto and listed by CISA on 14 Oct; no sign of access on our devices yet, log search unfinished. Done: management interfaces limited to the jump box. Open: 6 firewalls unpatched, Palo Alto lists no workaround and the jump box only limits who can reach the flaw, owner named, compromise search due 15 Oct 12:00. Next update: 15 Oct 08:00.
A fixed format makes each update comparable with the last one, and the unknown line stops later statements from outrunning the evidence. | Each brief with its time, saved as numbered versions in the incident record. | | 0 to 4 hours | Executive | Approve emergency change authority in writing, with limits: the actions allowed without the change board (apply a mitigation, apply the fixed build, restart the service), the systems covered, the people who may act, and an expiry 72 hours after hour zero. Removal of a safeguard stays with the normal board.
BOD 26-04 sets three days for a listed bug that needs forensic triage, which is shorter than a weekly board cycle. | Approval time, scope, named approvers and the expiry. | | 4 to 8 hours | Executive | Decide on downtime with three facts from the security lead: whether exploitation is confirmed, whether an unauthenticated network path exists, and whether a mitigation limits the reachable set to trusted networks. Ask the business owner for two numbers: cost per hour offline and the manual process that runs meanwhile. Then choose one: keep running behind the mitigation, run a reduced service, or take it offline until the fix is verified.
Only the executive can trade revenue for exposure, and the cost per hour is the number that lets the decision be reviewed instead of argued. | Decision, owner, time, cost per hour offline, the time of the next review. | | 4 to 24 hours | Executive | Ask counsel two questions in writing: which notification clocks apply to us if a compromise is found, and who makes the materiality determination if we are a US registrant. SEC Form 8-K Item 1.05 is due "within four business days after the registrant determines that it has experienced a material cybersecurity incident," and the determination "must be made without unreasonable delay after discovery of the incident." Put the determination meeting in the calendar with a time. The deadlines for each regime are on the communications playbook.
The 8-K clock starts at the determination, which is a decision of yours, and the instruction requires it "without unreasonable delay," so it cannot be put off to delay the filing. | Time counsel was asked, the written answer, the person who determines, the meeting time. | | 18 to 24 hours | Executive | Send the board brief 24 hours after hour zero, sooner if exploitation is confirmed on your own systems. It holds the five lines, the decisions taken with the time of each, the decisions that need the board, and what is unknown. Use dates and times, not words like recently.
A board that first hears of an incident from a regulator or a customer cannot help, and 24 hours leaves time for the first facts to firm up. | Version sent, time, recipients, board questions received. | | 24 to 48 hours | Executive | Review each risk acceptance as a signed record. Template, copy and fill: Asset: ____. Exposure (who can reach it, from where): ____. Flaw (CVE, authentication needed, impact): ____. Compensating control in place and tested on: ____. Reason it cannot be fixed now: ____. Owner: ____. Approver: ____. Expiry (date and hour): ____. Reject any acceptance for an unauthenticated flaw with total control and no compensating control. Set the expiry of an internet-reachable asset at 7 days or less, since 24 of the 33 dated bugs in the Disclosure Clock were exploited within that time.
An acceptance without an expiry becomes permanent, and the Clock shows the exposure is worst in the first week. | The signed record with the expiry, and the date it is reviewed. | | 24 to 72 hours | Executive | Ask for evidence, not status: the version output from the running service on each asset, and the security lead's triage outcome, one of CISA's three: no evidence of compromise, compromise suspected, compromise confirmed. Do not accept a change ticket as proof of a fix. If the outcome is suspected or confirmed, move the incident to the full incident response plan and call counsel the same hour.
Tickets close when work is requested, not when the flaw is gone, and the outcome decides whether any notification clock has started. | Evidence received, the outcome and the time you accepted it. | | 48 to 72 hours | Executive | Call the 72-hour review and assign each action item an owner and a date. Decide what to pre-approve for next time (emergency authority, a standing downtime rule), which end-of-life products to retire, and what to budget. Close with the hour-zero to verified-fix time as the number the board sees.
The decisions that were slow in hour 4 are the ones to settle while the facts are fresh. | Review notes, action items with owners and dates, the hour-zero to verified-fix time. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet When the vendor has no fix, the executive's decision stops being when to patch and becomes what to accept and for how long. Ask the vendor's executive contact for a written date and a written statement of what is known, then set a review at 24 hours. Accept only with a compensating control in place, and shorten the expiry in line with the evidence: if exploitation is confirmed, review every day. Cisco's ISE advisory shows how little a mitigation can be: it says "There are no workarounds that address this vulnerability. However, there is a mitigation," access lists, and calls such controls temporary. Some products will never get a fix. Cisco states that ISE release 3.0 "has reached End of Software Maintenance" and tells customers to migrate to a supported release, so for that release the choice is a migration budget or a retirement date. CISA's required action on a listed bug ends with "discontinue use of the product if mitigations are unavailable." The executive is the person who can make that call, and the call needs a number, the cost per hour of doing without the product, set against the exposure. ## CVE-2026-76460 worked through, hour by hour Cisco published an advisory for an authentication bypass in the API of Identity Services Engine on Wednesday 16 Sep 2026 at 16:00 GMT. ISE decides who joins the network, so the executive question was whether to change production at once on a flaw rated 10.0 that Cisco said was being exploited. The first two rows are the published facts. The rest are the executive decisions of this playbook with the clock time each falls due, counted from hour 0 at 16:00 GMT, so they are deadlines and not a record of what any company did. Each rests on a fact in the Cisco advisory or the CISA entry. | When | What happened | Source | | --- | --- | --- | | 16 Sep 2026, 16:00 GMT (hour 0) | Cisco publishes advisory cisco-sa-ISE-ABP-VNSW7Tn5 version 1.0, rated Critical with a base CVSS score of 10.0, and states: "The Cisco PSIRT is aware of active exploitation of this vulnerability." Fixed releases are 3.1 Patch 12, 3.2 Patch 11, 3.3 Patch 12, 3.4 Patch 7 and 3.5 Patch 4. Release 3.0 has reached End of Software Maintenance and has no fix. The only mitigation is infrastructure access control lists, and there is no workaround. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 16 Sep 2026 (hour 0) | CISA adds CVE-2026-76460 to the KEV catalog the same day, with a due date of 19 Sep 2026 and forensic triage marked Yes. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | | 16 Sep 2026, by 17:00 GMT (hour 1) | Decision: the executive writes hour 0 as 16:00 GMT, names the person who approves downtime and the person who accepts risk, and sets the bridge every 4 hours: 20:00, 00:00, 04:00 GMT and so on. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 16 Sep 2026, by 20:00 GMT (hour 4) | Decision: the first five-line brief is read and emergency change authority is signed. It covers the access lists and the fixed patches, and expires on Saturday 19 Sep at 16:00 GMT, the 72nd hour and the CISA due date. The patches are five, one per release train, and release 3.0 has none. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 17 Sep 2026, by 00:00 GMT (hour 8) | Decision: downtime call. The three facts are known from the advisory: exploitation is confirmed, no authentication is needed, and access lists limit who can reach the nodes. The executive chooses between running behind the lists, a reduced service and taking ISE offline, and writes down the cost per hour of doing without it. Cisco warns that evidence "may be removed or hidden," so a clean log alone does not close the question. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 17 Sep 2026, by 16:00 GMT (hour 24) | Decision: the board brief goes out, with the forensic triage outcome the KEV entry requires, and counsel has answered the notification-clock questions. Cisco strongly recommends re-imaging nodes where malicious activity is suspected, so the brief names a rebuild budget if a hit is found. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | | 18 Sep 2026, by 16:00 GMT (hour 48) | Decision: each open risk acceptance is signed on the template in step 7, with an expiry. Every ISE 3.0 node needs one, since 3.0 has no fix and its owner must name a migration date. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 19 Sep 2026, 16:00 GMT (hour 72) | The 72-hour review is held and CISA's due date falls, a Saturday, three days after the listing. The number reported to the board is the time from 16:00 GMT on 16 Sep to the verified fix. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | The advisory and the KEV entry both carried the facts for the first brief line at hour 0: exploitation confirmed by the vendor, and a due date three days out. Counted from 16:00 GMT on Wednesday, 72 hours end on Saturday 19 Sep at 16:00 GMT, so emergency change authority had to cover a weekend, and a weekly change board could not have met in time. The mitigation cost was the decision for hour 8: every admin station and integration left off the access lists would lose contact with the nodes. Five release trains needed five fixes, and one release had none, so the 72-hour review carried a retirement or migration item for ISE 3.0. ## Related pages - [The endpoint fleet playbook for the first 72 hours](/playbooks/endpoint-fleet/) - [The communications playbook for the first 72 hours](/playbooks/communications/) - [Dependency playbook: the first 72 hours when a library you ship has a critical flaw](/playbooks/dependency/) ## Sources - Cisco Security Advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-76460: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460 (read 19 Nov 2026) - CISA BOD 26-04, Prioritizing Security Updates Based on Risk: https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk (read 19 Nov 2026) - CISA BOD 26-04 Implementation Guidance, forensic triage steps: https://www.cisa.gov/news-events/directives/bod-26-04-implementation-guidance-prioritizing-security-updates-based-risk (read 19 Nov 2026) - SEC final rule, Cybersecurity Risk Management, Strategy, Governance, and Incident Disclosure, 88 FR 51896 (4 Aug 2023), Form 8-K Item 1.05: https://www.federalregister.gov/documents/full_text/html/2023/08/04/2023-16194.html (read 19 Nov 2026) - The Disclosure Clock, zerodayplaybook.com (33 bugs with a dated first exploitation): https://zerodayplaybook.com/clock/ (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/executive/ --- # The endpoint fleet playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Chrome 153.0.8010.36 fixed the exploited V8 flaw CVE-2026-87491 on 8 Sep 2026, yet a browser runs the old code until it relaunches. The steps set Chrome and Edge relaunch policies, find machines still running old code after the update landed, and chase apps that bundle their own Chromium. A browser fix is two events, not one: the new files arrive, and the running process restarts. Chrome applies an update only after a relaunch, and a person who keeps forty tabs open can postpone that for weeks. CVE-2026-87491 is an out-of-bounds write in V8, Chrome's JavaScript engine. The attacker needs only a crafted page, so a link, an ad or a compromised site is enough (CVSS vector AV:N, UI:R). The code runs inside the renderer sandbox, so reaching the machine usually takes a second flaw. Chromium rated the bug Medium, while CVSS 3.1 gives it 8.8; the Medium label did not stop Google from stating that an exploit exists in the wild. Nothing in the operating system patch tool owns this. Each Chromium browser takes the fix on its own schedule: Microsoft shipped Edge 152.0.4191.66 on 4 Sep 2026, four days before Chrome 153 reached the stable channel, and Google said Chrome 153 would "roll out over the coming days/weeks". Apps that embed Chromium, such as Electron and CEF applications, wait for their own vendors. Laptops sit off the network for days. The steps below cover what only an endpoint fleet has: versions that differ by vendor, a relaunch that users defer, and bundled runtimes. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Write the fixed-build list, one line per browser family. Chrome: 153.0.8010.36 on Linux, 153.0.8010.36 or .37 on Windows and Mac, 153.0.8010.36 on Android. Edge: Microsoft's Security Update Guide ties the CVE to Edge 152.0.4191.66 (Chromium 152.0.7977.83), and Edge 153.0.4234.32 and later builds on the 153 line carry it. Other Chromium browsers: open chrome://version and compare the Chromium version to 153.0.8010.36.
Chrome and Edge number their builds in different series, so comparing Edge 152.0.4191.66 to 153.0.8010.36 as plain numbers marks every patched Edge as old. | The list, with the page you took each build from and the time you read it. | | 0 to 4 hours | Platform engineering | Find machines that have the new files but still run the old code. Run this as SYSTEM so it sees every user's processes: foreach ($r in "$env:ProgramFiles\Google\Chrome\Application", "${env:ProgramFiles(x86)}\Google\Chrome\Application", "$env:LOCALAPPDATA\Google\Chrome\Application") { if (Test-Path $r) { $new = Get-ChildItem $r -Directory \| Where-Object Name -match '^\d+\.\d+\.\d+\.\d+$' \| Sort-Object { [version]$_.Name } \| Select-Object -Last 1; $old = Get-Process chrome -ErrorAction SilentlyContinue \| Where-Object { $_.Path -like "$r*" -and $_.StartTime -lt $new.CreationTime }; [pscustomobject]@{Host=$env:COMPUTERNAME; Installed=$new.Name; StaleProcesses=@($old).Count} } }. A process that started before the newest version folder was created is still running the previous build. A single machine shows the same state at chrome://settings/help as a Relaunch button. For installed versions only, use (Get-Item "$env:ProgramFiles\Google\Chrome\Application\chrome.exe").VersionInfo.ProductVersion, google-chrome --version on Linux, or the Browsers list in Chrome Browser Cloud Management. For installed Edge versions use (Get-Item "${env:ProgramFiles(x86)}\Microsoft\Edge\Application\msedge.exe").VersionInfo.ProductVersion on Windows, microsoft-edge --version on Linux, or edge://version on one machine.
Inventory tools report the installed version, which turns green the moment the files land, so a count of patched machines includes every browser still running vulnerable code. | One line per host: installed version, stale process count, run time. | | 0 to 4 hours | Platform engineering | Set the Chrome relaunch policies. RelaunchNotification = 2 (Required) and RelaunchNotificationPeriod = 14400000 (4 hours, in milliseconds). On Windows set both as REG_DWORD under HKLM\SOFTWARE\Policies\Google\Chrome; on Linux put {"RelaunchNotification": 2, "RelaunchNotificationPeriod": 14400000} in a file in /etc/opt/chrome/policies/managed/; on macOS set the same keys for com.google.Chrome; in Chrome Browser Cloud Management set the same policies in the console. Google's policy text: "If set to 'Required', a recurring warning will be shown to the user indicating that a browser relaunch will be forced once the notification period passes." The default period is one week (604800000) and the minimum is one hour (3600000). With 4 hours, the app menu changes at 1 hour 20 minutes, changes color at 2 hours 40 minutes, and the relaunch is forced at 4 hours. Leave RelaunchWindow unset: Google warns that "Setting this policy may delay application of software updates."
The default lets a user defer a pending update for seven days, which is 47 percent of the 15 days between Google's note of 8 Sep and the KEV due date of 23 Sep (14 days after the KEV listing). | The policy names and values, the group they apply to, and the time the first device reported them. | | 0 to 4 hours | Platform engineering | Set the same relaunch policies for Edge, which has the same names. Windows: REG_DWORD values RelaunchNotification = 2 and RelaunchNotificationPeriod = 14400000 (0x00DBBA00) under HKLM\SOFTWARE\Policies\Microsoft\Edge, or Group Policy path Administrative Templates/Microsoft Edge/Relaunch with the MSEdge.admx template. macOS uses the same key names. Microsoft: "If you set the policy to 'Required', a recurring warning prompts users that the browser will be restarted automatically as soon as a notification period passes." Edge refreshes the policy dynamically, so a running browser picks it up. Also set RelaunchFastIfOutdated to 7 (Chrome 137 and later, Edge 141 and later): when the running version has been outdated for more than that many days, the notification period drops to 2 hours. Both vendors' text for that policy calls Required "1" although the RelaunchNotification table says Required is 2, so set RelaunchNotification to 2.
Edge is a separate product with its own policy store, so the Chrome setting does nothing to it. | The Edge policy values and the groups; the RelaunchFastIfOutdated value as a standing setting. | | 4 to 24 hours | Platform engineering | Do not wait for the staged rollout. Push the current stable installer through your software tool: the Chrome Enterprise MSI on Windows, the package on macOS, and apt-get install --only-upgrade google-chrome-stable or dnf upgrade google-chrome-stable on Linux. For Edge, run the same push of the current Edge installer. A user can also start the update at Menu > Help > About Google Chrome. Then rerun the stale-process check from the second step.
The relaunch timer starts only once the update is staged on the machine, and Google said the staged rollout runs for days to weeks, so the policy alone does not bring the fix. | The push job name, the target group, and the count of machines that reported the new installed version. | | 4 to 24 hours | Platform engineering | Find Chromium that no browser update touches. On Windows run Get-ChildItem 'C:\Program Files','C:\Program Files (x86)',$env:LOCALAPPDATA -Recurse -Include libcef.dll,chrome_100_percent.pak -ErrorAction SilentlyContinue \| Select-Object -ExpandProperty DirectoryName -Unique. libcef.dll marks Chromium Embedded Framework applications and chrome_100_percent.pak marks Electron and Chromium builds. For an Electron app, read process.versions.chrome in the app's console. Then ask each app's vendor which release includes Chromium 153.0.8010.36 or later. Chrome for Android WebView follows the Chrome for Android update to 153.0.8010.36.
A Chrome update never reaches the Chromium copy inside another application, and the app stays exposed until its vendor ships a build. | App, folder, bundled Chromium version, vendor reply and its date. | | 24 to 72 hours | Platform engineering | Block laptops that stay off the network until they update. In Intune create a custom compliance policy for Windows with a discovery script that prints {"ChromeCurrent": true} or false from $v = (Get-Item "$env:ProgramFiles\Google\Chrome\Application\chrome.exe" -ErrorAction SilentlyContinue).VersionInfo.ProductVersion; ($v -and [version]$v -ge [version]'153.0.8010.36'), and a rule on ChromeCurrent with the operator IsEquals, data type Boolean and operand true. Require a compliant device in Conditional Access for corporate web apps. Add the same pair for Edge, reading msedge.exe from "${env:ProgramFiles(x86)}\Microsoft\Edge\Application". Its floor is 152.0.4191.66 on the 152 line, or 153.0.4234.32 on the 153 line and later (Microsoft's release notes of 4 and 10 Sep 2026); compare each build against the floor of its own line.
A laptop that is not on the network cannot receive the push, and the relaunch timer does not start until the update arrives. | The policy name, the group it applies to, and the number of devices it marked non-compliant at hours 24, 48 and 72. | | 24 to 72 hours | Security lead | Report two numbers every 12 hours: the share of machines on a fixed installed version, and the share with no stale process. Stop at the second. Then close: list every exception, whether a machine, an embedded app or a vendor without a date, with an owner and a date, and give the people exposed on an old build after 8 Sep the same notice as everyone else. Google, Microsoft and CISA publish no indicators of compromise for this flaw and the KEV entry marks forensicTriage No, so there is no hunt to run; the record of who ran old code, and when, is the evidence.
The installed-version share reaches 100 percent while browsers still run old code, so only the second number says the exposure has ended. | The two percentages at each report time, and the exceptions list with owner and date. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet Google lists no workaround or settings change for this flaw, and CISA's required action is to apply the vendor's mitigations or stop using the product. For a Chromium build with no fix yet, such as an embedded browser whose vendor gives no date, there are two controls and both are our suggestion, not Google's. First, restrict what the app can open: if it renders only your own pages, the attacker has no route to it; if it opens any URL, add a URLBlocklist or an allow list. The cost is blocked sites. Second, turn off the JavaScript JIT: DefaultJavaScriptJitSetting = 2 in Chrome, and the same policy exists for Edge. Google says this "may render web content more slowly, and may also disable parts of JavaScript including WebAssembly", and it helps only if this flaw sits in the JIT compiler, which Google has not said. Treat it as a stopgap for people who face untrusted pages all day. Take an application out of use when it renders pages from the open web, its vendor has no fixed build and no date, and people cannot work without a different route. Leave it running when it shows only first-party content and its vendor has given a date inside the KEV window of 14 days. Remember the sandbox: the code runs inside the renderer, so an unpatched browser alone does not hand over the machine, and the real escalation needs a second flaw that no source ties to this one. ## CVE-2026-87491 worked through, day by day Google's Chrome 153 release note of 8 Sep 2026 listed 230 security fixes. One was CVE-2026-87491, an out-of-bounds write in V8 that Google rated Medium and described as having an exploit in the wild. Microsoft's Edge notes show the fix arriving earlier in Edge than in Chrome's public note, and CISA listed the CVE the day after Chrome 153 went out. The sources give dates, and hours only for the Microsoft record, so the table runs in days. | When | What happened | Source | | --- | --- | --- | | 6 Aug 2026 | Jihyeon Jeong of Compsec Lab, Seoul National University reports the bug as Chromium issue 543557673. Google pays $2,500. | [source](https://chromereleases.googleblog.com/2026/09/stable-channel-update-for-desktop_0808145027.html) | | 4 Sep 2026 | Microsoft releases Edge 152.0.4191.66. Its release note says "The Chromium team reported that CVE-2026-87491 has an exploit in the wild, and this update contains a fix for it." | [source](https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security) | | 8 Sep 2026 | Chrome 153.0.8010.36 (Linux) and 153.0.8010.36/.37 (Windows, Mac) reach the stable channel. The note says the release "will roll out over the coming days/weeks" and that Google is aware an exploit for CVE-2026-87491 exists in the wild. The bug is listed as Medium. | [source](https://chromereleases.googleblog.com/2026/09/stable-channel-update-for-desktop_0808145027.html) | | 9 Sep 2026 | CISA adds CVE-2026-87491 to KEV with forensicTriage No and a due date of 23 Sep 2026. The entry says it could affect Chrome, Edge and Opera. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | | 10 Sep 2026 | Microsoft releases Edge 153.0.4234.32 for Stable. | [source](https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security) | | 11 Sep 2026, 21:56 (UTC minus 7) | Microsoft's Security Update Guide record for the CVE is released. It says Edge "ingests Chromium" and repeats that Google is aware of an exploit in the wild. | [source](https://api.msrc.microsoft.com/sug/v2.0/en-US/vulnerability/CVE-2026-87491) | | 23 Sep 2026 | KEV due date for CVE-2026-87491: 14 days after the CISA listing of 9 Sep and 15 days after the Chrome note of 8 Sep. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | On the default policy, a laptop that stages the update on day 3 (11 Sep) is forced to relaunch on day 10 (18 Sep), 168 hours later and 5 days before the KEV due date. A laptop that stages it on day 12 passes the due date before the timer ends. With RelaunchNotificationPeriod at 14400000, the relaunch is forced 4 hours after the update stages. The push in the fifth step moves staging from days or weeks to hours 4 to 24 for machines on the network, so the fleet finishes online machines by about hour 28 and spends the remaining time on laptops and bundled apps. An Edge fleet that took the 4 Sep build had the fix 4 days before Google's note existed. ## Related pages - [Dependency playbook: the first 72 hours when a library you ship has a critical flaw](/playbooks/dependency/) - [The executive playbook for the first 72 hours](/playbooks/executive/) - [The network edge appliance playbook for the first 72 hours](/playbooks/network-edge-appliance/) ## Sources - Chrome Releases, Stable Channel Update for Desktop, 8 Sep 2026: https://chromereleases.googleblog.com/2026/09/stable-channel-update-for-desktop_0808145027.html (read 19 Nov 2026) - Chrome Enterprise policy list data (RelaunchNotification, RelaunchNotificationPeriod, RelaunchWindow, RelaunchFastIfOutdated, DefaultJavaScriptJitSetting): https://chromeenterprise.google/static/json/policy_templates_en-US.json (read 19 Nov 2026) - Microsoft Edge policy documentation, RelaunchNotification: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-policies/relaunchnotification (read 19 Nov 2026) - Microsoft Edge policy documentation, RelaunchNotificationPeriod: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-policies/relaunchnotificationperiod (read 19 Nov 2026) - Microsoft Edge policy documentation, RelaunchFastIfOutdated: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-policies/relaunchfastifoutdated (read 19 Nov 2026) - Microsoft Edge security release notes: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security (read 19 Nov 2026) - Microsoft Security Update Guide record for CVE-2026-87491: https://api.msrc.microsoft.com/sug/v2.0/en-US/vulnerability/CVE-2026-87491 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities feed, catalog version 2026.11.18: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/endpoint-fleet/ --- # Dependency playbook: the first 72 hours when a library you ship has a critical flaw Playbook, sources read 19 Nov 2026 **The short answer.** When a library or runtime your application depends on gets a critical flaw, the owning team has 72 hours to find every copy, prove which paths reach it, force the fixed version through transitive pins, and block the old range in CI. This page gives the queries and a Brotli case. A dependency flaw is different from a flaw in a product you installed. Nobody on your team chose the vulnerable code, and it often does not appear in a manifest: the Brotli decoder behind CVE-2026-111497 is compiled into the Node.js binary, so no package file lists it. Attackers exploit these flaws through the data your application already accepts. In the Brotli case that was a request body sent with Content-Encoding br, or a compressed response fetched from a URL the attacker influences. What the attacker gains is code execution inside your service, with its credentials and network position. The page is for the team that owns an application. It assumes you will not be told which of your services is affected. The hard parts are three. A fix released for the parent package may leave your copy untouched. The version string may not tell patched from vulnerable builds. And the owner has to prove, service by service, whether untrusted input reaches the decoder. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Application owner | Write down four facts from the advisory before you search: the package coordinates as a package URL (for example pkg:pypi/brotli), the affected range, the fixed versions, and the trigger condition. For CVE-2026-111497 these are Brotli 1.0.3 up to but excluding 1.0.10, 1.1.0 up to but excluding 1.1.1 and 1.2.0 up to but excluding 1.2.1, fixed in 1.0.10, 1.1.1 and 1.2.1, and Node.js 22.x before 22.23.4, 24.x before 24.22.1 and 26.x before 26.12.1. Node.js lines 20.x, 23.x and 25.x are end-of-life and get no fix. The trigger is decompressing Brotli data an attacker can influence. Also note that the package may be bundled in a runtime rather than installed as a package: here Node.js carries its own copy.
Every later query needs the exact range and the bundling fact, and a search for the package name alone misses the copy inside the runtime. | A ticket header with the CVE, the coordinates, the ranges, the fixed versions, the trigger and the time you started. | | 0 to 4 hours | Application owner | Query your SBOM store for both the package and any runtime that bundles it. For CycloneDX files on disk: jq -r '.components[]? \| select(.name\|test("brotli";"i")) \| [.name, .version, .purl] \| @tsv' bom.json, run in a loop over every file. For Dependency-Track: curl -s -H "X-Api-Key: $KEY" "$DT/api/v1/component/identity?name=brotli". Repeat both with name node and keep the versions below 22.23.4, 24.22.1 and 26.12.1. A generated SBOM for an image is also queryable: syft -o cyclonedx-json \| jq ... . Expect the runtime query to carry the real hits, because the bundled decoder is not a component of its own.
The SBOM answers 'where is it' for hundreds of services in minutes, and the second query covers the copy the first cannot see. | The query text, the date and time you ran it, the SBOM store version, and the hit list with service, component and version. | | 0 to 4 hours | Application owner | For services without a current SBOM, ask the package manager for the dependency path. npm ls --all \| grep -i brotli and, for the inbound paths, npm ls express body-parser. Python: pip show brotli brotlicffi and pipdeptree -r -p brotli. Maven: mvn dependency:tree -Dincludes=org.brotli. Gradle: gradle dependencies --configuration runtimeClasspath \| grep -i brotli. Go: go mod graph \| grep -i brotli and go version -m ./bin/app \| grep -i brotli. Ruby: bundle info brotli. Runtime: node --version on hosts and CI runners, docker run --rm node --version for images, and ldd $(command -v node) \| grep -i brotli to see whether Node.js links the system library, then dpkg -l \| grep -i brotli, rpm -q brotli or apk info -v \| grep brotli. Do not test process.versions.brotli: ELLIO reports it shows 1.2.0 on patched and vulnerable builds alike.
The pipdeptree and npm ls output show which parent pulls the vulnerable copy in, which decides whether you can fix it yourself or must wait for another team's release. | One line per service: the ecosystem, the command, the version found, and the parent package for each transitive hit. | | 0 to 4 hours | Application owner | Decide reachability for each hit, using the three paths into the decoder. Inbound bodies: Express 5, body-parser 2.x and Koa with @koa/bodyparser inflate Brotli request bodies before route handlers run, while a bare node:http server does not. Outbound responses: built-in fetch(), undici, axios, got and SuperAgent decode Brotli responses, so webhooks, link previews and crawlers that fetch URLs a user supplies count. Direct calls: grep -rn "brotliDecompress\\|createBrotliDecompress" . in your code and in node_modules. Then test one staging service: POST a corrupt body (not Brotli) with the header Content-Encoding: br. A 400 from the framework before your handler runs usually means a decoder ran on the body.
Reachability sets the order of work: an internet-facing service that decodes request bodies goes first, a batch job that never touches untrusted compressed data goes last. | Per service: reachable yes, no or unknown; which of the three paths; whether the proxy forwards Content-Encoding br; and who is on call. | | 4 to 24 hours | Application owner | Move each reachable service to the fixed version, including the dependencies you do not declare. For a runtime, change the pinned base: a Dockerfile pinned to node:24.22.0 or node:22.23.3 stays vulnerable until the pin moves. For a transitive package, force it from your own manifest: npm overrides in package.json, yarn resolutions, a Maven dependencyManagement entry, a pip constraints file passed with -c, or go get @. Python Brotli moves to 1.2.1 and brotlicffi to 1.2.1.0. Open one pull request per repository and run the suite. A Node.js upgrade fixes only the copy Node.js bundles, so rerun step 3 after the change.
A pin in your own manifest is the only fix that does not wait for another maintainer, and the rerun proves the pin held. | The pull request link, the pin you added, the before and after version, and the CI run. | | 4 to 24 hours | Application owner | Give platform engineering the smoke tests and the rollback criteria before they deploy. For a decoder fix, the tests are: a valid Brotli request body decodes to the same content as before; a Brotli response from a stub upstream decodes through your fetch client; and a corrupt body is rejected with a 400 and does not crash the worker. Rollback criteria: crash-loop restarts, or a jump in 400 responses on requests with Content-Encoding br beyond what the corrupt-body test produces. A patched decoder rejects a malicious body with ERR__ERROR_FORMAT_BLOCK_SWITCH (decoder error code -17), so that string in logs is expected after the change.
The owner knows which behavior the patch must not change; without written criteria the rollout either stalls or ships blind. | The test list, the pass or fail result against staging, and the threshold that triggers rollback. | | 4 to 24 hours | Application owner | For services that cannot move yet, ask platform engineering for the inbound stopgap at the proxy and confirm it with a request. In nginx: if ($http_content_encoding ~* "\bbr\b") { return 415; } in the server block, then curl -s -o /dev/null -w '%{http_code}\n' -H 'Content-Encoding: br' --data-binary @bad.bin https://staging.example/api should print 415. Aikido and ELLIO give this stopgap because the Node.js advisory lists no workaround. It does not cover Brotli responses your own service fetches. For those, disable or restrict features that fetch user-supplied URLs until the upgrade ships.
The stopgap closes the inbound path in minutes, while the outbound path needs a feature decision that only the owner can make. | The proxy rule and its commit, the test request and its status code, the feature flags you turned off, and the end date of the exception. | | 24 to 72 hours | Application owner | Check the built artifacts, not the repositories. For every production image run docker run --rm node --version and compare with 22.23.4, 24.22.1 and 26.12.1, and rerun the SBOM and pip show queries against what is deployed. Do not accept a scanner's Brotli finding or silence as the proof for Node.js: the version printed by the image is the proof.
A merged pull request does not change a running image, and a fixed repository can still ship through a stale tag. | The image digest, the version printed, the host or pod, and the time of the check. | | 24 to 72 hours | Application owner | Look back through the exposure window for attempts and for crashes. Search application logs for ERR__ERROR_FORMAT_BLOCK_SWITCH, which marks a failed attack against a patched build. On vulnerable builds look for repeated worker crashes, restarts and requests with Content-Encoding br just before them, and check /var/lib/systemd/coredump: ELLIO notes a crash core file can hold credentials from other users' sessions, so restrict who can read it and how long you keep it. Some other decoders (Python Brotli 1.2.0, brotlicffi 1.2.0.0) return no bytes, never finish and raise no error, so look for stuck requests with CPU at 100 percent too. If a vulnerable, internet-reachable service saw this traffic or unexplained crashes, treat the host as possibly compromised and start your incident process.
Detection comes after the fix so the fix is not delayed, but the exposure window is closed only when the logs are read. | The searches you ran with their time ranges, the counts, and either 'no indicators' or the host list handed to incident response. | | 24 to 72 hours | Application owner | Make the old range fail the build. Add a CI step that reads the runtime version and exits nonzero below the fixed release of its line: v=$(node -p process.versions.node); case $v in 22.*) m=22.23.4;; 24.*) m=24.22.1;; 26.*) m=26.12.1;; *) exit 1;; esac; printf '%s\n%s\n' $m $v \| sort -V -C \|\| { echo old Node.js; exit 1; }. Add a lock-file check for the other packages, for example trivy fs --exit-code 1 --severity CRITICAL . on the repository. Keep the override from step 5 until the parent package ships the fix on its own.
Without a gate, the next base-image bump or a lock-file regeneration can bring the old range back. | The CI job name, the commit that added it, and the date you plan to remove the temporary override. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet Three kinds of dependency have no fix when you need one. The first is an end-of-life runtime line: Node.js 20.x, 23.x and 25.x get no release for CVE-2026-111497, so the only complete answer is a move to 22, 24 or 26. The second is a copy inside software you cannot rebuild: NVD lists many Siemens products as embedding Brotli, and the fix arrives when the vendor ships it. The third is a library with no release yet. The Brotli fix sat in the public master branch of google/brotli from 27 Feb 2026, eight months before a release carried it, so a snapshot build from the fix commit was possible. Its cost is a build nobody has tested and a copy you now maintain. For services that stay on a vulnerable copy, the mitigations are the proxy rule for inbound Content-Encoding br, the platform edge rules where they exist (Vercel added them on 21 Oct 2026 for inbound requests only), and removing the features that fetch user-supplied URLs. Each costs a class of legitimate traffic or a feature. Take the service offline, or the feature off, when it decodes attacker-influenced data on a path you cannot filter and the copy cannot be replaced: CISA's KEV action for this CVE is to apply the vendor mitigations or stop using the product if none are available. A batch job that never reads untrusted compressed input can stay up with the exception recorded and dated. ## CVE-2026-111497 worked through, hour by hour The Brotli decoder flaw is a dependency case in its purest form. The vulnerable code was Google's Brotli, shipped inside Node.js, inside Python packages and inside other products. CISA lists it in the KEV catalog as Google Brotli Unchecked Return Value Vulnerability. NVD rates it CVSS 9.8. The Node.js project named the reporters and the fixed releases on 21 Oct 2026. The rows below start with the commit that carried the fix and end with the CISA due date. | When | What happened | Source | | --- | --- | --- | | 27 Feb 2026 | The fix is committed to the public master branch of google/brotli with no release and no advisory, 236 days before the Node.js release. | [source](https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/) | | 14 Oct 2026 | Node.js maintainers post a pre-alert on oss-security: new releases of Node.js 22, 24 and 26 on or shortly after 21 Oct, the highest severity being HIGH. It gives seven days to take the inventory of Node.js versions in steps 1 to 3. | [source](http://www.openwall.com/lists/oss-security/2026/10/14/10) | | 21 Oct 2026 | Node.js releases 22.23.4, 24.22.1 and 26.12.1 and publishes the advisory. Brotli 1.2.1 ships the same day, and 1.0.10 and 1.1.1 carry the fix too. Vercel adds firewall rules for inbound requests. | [source](https://nodejs.org/en/blog/vulnerability/october-2026-security-releases) | | 22 Oct 2026 | NVD publishes the record and GitHub lists GHSA-9v8h-qqgm-4m35. A first public proof of concept appears. Aikido says it reached code execution only against a deliberately weakened test build. | [source](https://www.aikido.dev/blog/node2shell-nodejs-brotli-rce-cve-2026-111497) | | 23 Oct 2026 | A post on oss-security points out that the flaw is in Google Brotli and reaches every consumer, and that Ubuntu rates it medium. | [source](http://www.openwall.com/lists/oss-security/2026/10/23/25) | | 25 Oct 2026 | ELLIO reports a scanner named node2shell-check mapping hosts that decompress Brotli, and reproduces code execution on default Node.js 24.22.0 on x86-64. | [source](https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/) | | 26 Oct 2026, 15:08 UTC | ELLIO sensors record a Brotli-coded request body of 23,069,983 bytes, the first exploit traffic it reports. | [source](https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/) | | 27 Oct 2026 | CISA adds the CVE to the KEV catalog with a due date of 10 Nov 2026. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-111497) | Count from the advisory on 21 Oct. The 72-hour window ends on 24 Oct, two days before the first exploit traffic ELLIO reports on 26 Oct and three days before the KEV entry. The pre-alert of 14 Oct named the release lines and the date, so the inventory of Node.js versions in use on lines 22, 24 and 26 could finish a week ahead of the first traffic. A team that relied on the SBOM alone would have missed the Node.js-bundled copy and the Dockerfile pins to node:24.22.0, which stay vulnerable after a base-image refresh. The NVD record lists Siemens products as embedding Brotli: for those, steps 5 to 7 end with a vendor ticket and the proxy rule, and the KEV due date of 10 Nov 2026 is the date the exception must close. ## Related pages - [The network edge appliance playbook for the first 72 hours](/playbooks/network-edge-appliance/) - [The endpoint fleet playbook for the first 72 hours](/playbooks/endpoint-fleet/) - [Cisco ISE network access control playbook: the first 72 hours after a critical flaw](/playbooks/network-access-control/) ## Sources - Node.js security releases, 21 Oct 2026 (vendor advisory): https://nodejs.org/en/blog/vulnerability/october-2026-security-releases (read 19 Nov 2026) - CISA KEV entry for CVE-2026-111497: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-111497 (read 19 Nov 2026) - ELLIO research on Node2Shell exploitation and affected builds: https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/ (read 19 Nov 2026) - Aikido analysis and fix guidance: https://www.aikido.dev/blog/node2shell-nodejs-brotli-rce-cve-2026-111497 (read 19 Nov 2026) - Node.js security release pre-alert on oss-security, 14 Oct 2026: http://www.openwall.com/lists/oss-security/2026/10/14/10 (read 19 Nov 2026) - oss-security post, 23 Oct 2026: http://www.openwall.com/lists/oss-security/2026/10/23/25 (read 19 Nov 2026) - Google Brotli 1.2.1 release: https://github.com/google/brotli/releases/tag/v1.2.1 (read 19 Nov 2026) - NVD record for CVE-2026-111497: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-111497 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/dependency/ --- # The network edge appliance playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Firewalls, VPN gateways and other internet-facing appliances sit in the traffic path, and CISA added the PAN-OS flaw CVE-2026-0322 to KEV on the day Palo Alto disclosed it. This playbook gives the hour-by-hour order, the fixed builds for all five PAN-OS trains, the SonicOS firmware, the exposure checks and the no-fix branch. An edge appliance is a closed box that sits where traffic enters. Flaws in it are usually exploited through the management interface or the VPN portal. In PAN-OS CVE-2026-0322 the attacker needs network reach to the management interface and a valid high-privilege account (CVSS 4.0 privileges required: high, base score 8.6), and a successful attack runs code as root. The attacker then holds the device that sees, decrypts and routes your traffic, plus the credentials, certificates and VPN sessions stored on it. No EDR agent runs on most of these boxes, so you cannot rely on endpoint telemetry to tell you what happened. Patching is harder than on a server. The device is in the traffic path, so every upgrade is an outage unless you run a high-availability pair. Hotfix builds are cut per release train, so the fixed build depends on your minor release: PAN-OS CVE-2026-0322 has fixes in five trains and 20 builds, listed below. The exploit often needs a condition you can read from the device, such as a reachable management interface or SAML on the SSLVPN, so the first hours go to finding which boxes meet it. This page also covers other internet-facing services, because the same order applies: cut reachability, patch, then look for earlier access. PAN-OS CVE-2026-0322 fixed builds, one line per release train (affected range, then the build to upgrade to). Cloud NGFW and Prisma Access are not affected, and every older unsupported PAN-OS version is affected and must move to a supported fixed version. PAN-OS 12.2. 12.2.0 to 12.2.3: upgrade to 12.2.4 or later. PAN-OS 12.1. 12.1.8 up to, not including, 12.1.10-h2: upgrade to 12.1.10-h2 or later; 12.1.5 up to, not including, 12.1.7-h8: upgrade to 12.1.7-h8, or 12.1.10-h2 or later; 12.1.2 up to, not including, 12.1.4-h12: upgrade to 12.1.4-h12, or 12.1.10-h2 or later. PAN-OS 11.2. 11.2.14 to 11.2.15: upgrade to 11.2.16 or later; 11.2.11 up to, not including, 11.2.13-h3: upgrade to 11.2.13-h3, or 11.2.16 or later; 11.2.8 up to, not including, 11.2.10-h15: upgrade to 11.2.10-h15, or 11.2.16 or later; 11.2.5 up to, not including, 11.2.7-h21: upgrade to 11.2.7-h21, or 11.2.16 or later; 11.2.0 up to, not including, 11.2.4-h22: upgrade to 11.2.4-h22, or 11.2.16 or later. PAN-OS 11.1. 11.1.14 up to, not including, 11.1.16-h3: upgrade to 11.1.16-h3 or later; 11.1.11 up to, not including, 11.1.13-h15: upgrade to 11.1.13-h15, or 11.1.16-h3 or later; 11.1.8 up to, not including, 11.1.10-h35: upgrade to 11.1.10-h35, or 11.1.16-h3 or later; 11.1.7 up to, not including, 11.1.7-h11: upgrade to 11.1.7-h11, or 11.1.16-h3 or later; 11.1.5 up to, not including, 11.1.6-h39: upgrade to 11.1.6-h39, or 11.1.16-h3 or later; 11.1.0 up to, not including, 11.1.4-h37: upgrade to 11.1.4-h37, or 11.1.16-h3 or later. PAN-OS 10.2. 10.2.17 up to, not including, 10.2.18-h12: upgrade to 10.2.18-h12 or later; 10.2.14 up to, not including, 10.2.16-h11: upgrade to 10.2.16-h11, or 10.2.18-h12 or later; 10.2.11 up to, not including, 10.2.13-h26: upgrade to 10.2.13-h26, or 10.2.18-h12 or later; 10.2.8 up to, not including, 10.2.10-h41: upgrade to 10.2.10-h41, or 10.2.18-h12 or later; 10.2.0 up to, not including, 10.2.7-h38: upgrade to 10.2.7-h38, or 10.2.18-h12 or later. SonicOS CVE-2026-19868 (SonicWall advisory SNWLID-2026-0014). Gen7 firmware 7.3.3-7015 and older: upgrade to 7.3.4-7013 or higher. Gen8 firmware 8.2.2-8015 and older: upgrade to 8.2.3-8015 or higher. NVD's data uses slightly different cut-offs, so follow SonicWall's figures: 8.2.2-8015 is still affected. The flaw is a SAML bypass in the SSLVPN, exploited per CISA's KEV entry; the dates are in the case table. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | List every appliance of the vendor, its model family and software version. PAN-OS (CVE-2026-0322): run `show system info` and read model and sw-version, or read Dashboard > General Information; check Panorama and each firewall it manages. SonicOS (CVE-2026-19868): read the firmware version in the management interface for every firewall of a pair and every NSv. Match each version to the vendor's fixed-build table for its own release train.
The fixed build differs per train, so a version that looks recent can still be below its train's fix (PAN-OS 11.2.13-h2 and 11.2.15 are both affected). | One row per device: name, model, version, fixed build for its train, affected yes or no. | | 0 to 1 hour | Security lead | Check the exposure conditions the advisory names. PAN-OS CVE-2026-0322: read Device > Setup > Interfaces > Management Interface Settings > Permitted IP Addresses, and Network > Network Profiles > Interface Mgmt for any profile that allows HTTPS or SSH on a data-plane interface. Then test from a host outside your network, such as a cloud instance, with `nmap -Pn -p 22,443 ` and `curl -sk -m 5 -o /dev/null -w "%{http_code}\n" https:///`; any answer means the interface is reachable from the internet. SonicOS (CVE-2026-19868) has two preconditions. First, SAML is the login method for SSLVPN users: review the SSLVPN user authentication settings in the management interface. Second, at least one LDAP, RADIUS or TACACS+ server is active: open Device > Users > Settings > Authentication, then Configure RADIUS > Settings > RADIUS Servers, Configure LDAP > Settings > LDAP Servers and Configure TACACS+ > Settings > TACACS+ Servers; any server listed and not disabled counts. If either half fails, the bypass does not apply to that firewall. The attacker also needs valid credentials for one of those alternate methods, so list who holds accounts in those directories and confirm whether the SSLVPN portal answers from an untrusted network.
A flaw with a stated precondition is not the same urgency on every box; the answer sorts devices into reachable now and not reachable. | Per device: management address reachable from the internet yes or no, and the precondition result. | | 0 to 4 hours | Platform engineering | Cut reachability before the patch. For PAN-OS, set Permitted IP Addresses to one jump box, which is the only system allowed to reach the management interface. For SSLVPN flaws, restrict the portal to known source ranges only if remote staff can work without it. Test from an untrusted network that the management address no longer answers.
This is the one action that fits inside the first hours without an outage window; the CVSS 4.0 requirement of a high-privilege account means it removes the network path an attacker needs. | The change ticket, the before and after Permitted IP Addresses, and the outside test result. | | 0 to 4 hours | Security lead | Pull forensic evidence before any upgrade. On PAN-OS export device state (Device > Setup > Operations > Export device state) and save System, Config and Authentication logs from Monitor > Logs. Save the configuration audit and the administrator list (Device > Administrators). Send logs to storage the appliance cannot write to.
An upgrade restarts the device and can overwrite logs, and CISA marked forensic triage as required for the PAN-OS KEV entry. | File names, hashes and the time of export, kept with the incident record. | | 0 to 4 hours | Security lead | Search the exported logs and the firewall traffic logs for the indicator addresses the vendor published. For CVE-2026-0322 these are 66[.]42[.]72[.]14, 208[.]83[.]239[.]114 and 149[.]248[.]47[.]234. Palo Alto published no file or log signatures, so also list every administrator login to the management interface in the exposure window and check each against a person.
The vendor gives only three addresses, so an empty search proves little; the administrator login list covers what the addresses miss. | Search period, queries used, hit count (zero is a result), and the admin logins that nobody could explain. | | 4 to 24 hours | Platform engineering | Upgrade to the fixed build for the train, in a window within the day. PAN-OS (CVE-2026-0322), by release train: 12.2 to 12.2.4; 12.1 to 12.1.4-h12, 12.1.7-h8 or 12.1.10-h2; 11.2 to 11.2.4-h22, 11.2.7-h21, 11.2.10-h15, 11.2.13-h3 or 11.2.16; 11.1 to 11.1.4-h37, 11.1.6-h39, 11.1.7-h11, 11.1.10-h35, 11.1.13-h15 or 11.1.16-h3; 10.2 to 10.2.7-h38, 10.2.10-h41, 10.2.13-h26, 10.2.16-h11 or 10.2.18-h12 (each or later in its sub-train, per Palo Alto). SonicOS (CVE-2026-19868): Gen7 to 7.3.4-7013 or higher, Gen8 to 8.2.3-8015 or higher. In an HA pair, suspend the passive node, upgrade and verify it, fail over, then upgrade the other, and take a configuration backup first. On PAN-OS run `show system info` after each node and confirm sw-version equals the fixed build.
Pairs let you patch with no outage, and the failover test proves traffic works on the new build before the second node changes. | Per node: build before, build after, time, and the traffic test result. | | 4 to 24 hours | Platform engineering | For devices you cannot upgrade today, keep the reachability cut in place and apply the vendor's partial coverage if it exists. PAN-OS (CVE-2026-0322): Threat IDs 510046 and 510047 (content version 9159-10287 or later) need management traffic routed through a data-plane interface with decryption and threat prevention; treat them as partial cover only. SonicOS (CVE-2026-19868): the workaround is to disable the LDAP, RADIUS and TACACS+ servers, which removes the alternate login method the bypass needs; use it only where no one needs those servers.
A device that stays on a vulnerable build needs a named reason and an end date, not silence. | Device, reason it was not patched, the interim control, and the date it will be patched. | | 24 to 72 hours | Security lead | PAN-OS (CVE-2026-0322): hunt for persistence on every device that was reachable before the cut. Open Device > Config Audit and compare the running configuration with the last backup taken before the device was first reachable from an untrusted network, not the 14 Oct 2026 disclosure date, because exploitation can predate an advisory; with no such backup, use the oldest one you hold and treat each difference as unexplained until a person owns it. Open Device > Administrators and match every account and role to a person. Read Monitor > Logs > Configuration and System for administrator logins, and for changes made by accounts nobody can explain. Check Device > Certificate Management > Certificates for certificates you did not install, and Device > Server Profiles and Device > Authentication Profile for changed RADIUS, LDAP or SAML servers. Check Network > Virtual Routers and Policies for routes, NAT and rules added since that last clean backup. Because the flaw runs code as root, a device with any unexplained change is rebuilt, not cleaned. SonicOS (CVE-2026-19868): compare SSLVPN login records with the SAML provider's sign-in log: a user on the firewall with no matching SAML sign-in came in through the alternate method. SonicWall names no log event for this bypass, so export the SSLVPN login events from the firewall event log or the syslog server for the exposure window into firewall-sslvpn-users.txt (one username per line) and the SAML sign-ins for the same window from the identity provider into idp-saml-signins.txt, then run comm -23 <(sort -u firewall-sslvpn-users.txt) <(sort -u idp-saml-signins.txt). Every name it prints logged into SSLVPN without a SAML sign-in and gets an owner's explanation, or its credentials are rotated and its sessions ended.
Patching does not remove access gained before it, and root access lets the attacker hide changes on the box itself, so the PAN-OS audit and administrator views are where a change shows. | Diff of config against the last known good backup, with each difference explained or escalated. | | 24 to 72 hours | Platform engineering | If any device was reachable and cannot be cleared, rotate what it holds: administrator and service passwords, LDAP and RADIUS bind credentials, VPN pre-shared keys, SAML signing and device certificates, and force all VPN sessions to end.
Credentials on the device are readable by anyone with root, so a patch alone leaves them exposed. | Each secret rotated, the date, and the owner who confirmed the dependent system still works. | | 24 to 72 hours | Security lead | Rescan from outside the network for the vulnerable version banner and the management port, and confirm each device meets its fixed build. Close the temporary jump-box rule only when the owner agrees, and keep it if no one needs the management interface from elsewhere.
The outside view is the only check that includes devices missing from the inventory. | Scan date, devices seen, any not in the inventory, and the sign-off. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet Some edge flaws ship with no usable workaround. Palo Alto states that no known workaround exists for CVE-2026-0322, so the only options are to shrink who can reach the interface or to filter the traffic. A jump box that is the only system allowed to reach the management interface costs you remote administration from anywhere else, and any script that manages the device from another address stops working. Palo Alto rates that case 8.5 instead of 8.6 because the attack vector becomes adjacent, so it lowers risk a little and does not remove it. Threat IDs 510046 and 510047 cost decryption load and certificate work, and Palo Alto calls their coverage limited. Take the device offline from the internet when three things are true: the management interface or portal is reachable from untrusted networks, the vendor or CISA reports exploitation, and the only mitigation is partial. Switching off the management interface does not stop the traffic path, so for management flaws you can usually cut reachability without ending service. For a VPN portal flaw you cannot: SonicWall's workaround for CVE-2026-19868 (updated 9 Nov 2026) is to disable or delete every LDAP, RADIUS and TACACS+ server on the firewall, under Device > Users > Settings > Authentication, which stops anything signing in through them. Decide by listing what depends on those servers before you disable them. ## CVE-2026-0322 worked through, day by day Palo Alto Networks disclosed CVE-2026-0322 on 14 Oct 2026: code injection (CWE-94) in the PAN-OS management interface, CVSS 4.0 base score 8.6 (NVD's own CVSS 3.1 rating is 7.2), exploitable by an attacker who can reach the interface and holds a high-privilege account. The vendor said very limited exploitation had been observed. The advisory and KEV give dates, not hours, so the table runs in days and is in date order. The SonicOS rows for CVE-2026-19868 are marked and fall in the same order. | When | What happened | Source | | --- | --- | --- | | 14 Oct 2026 | Palo Alto publishes the advisory with fixed builds in five release trains, exploit maturity ATTACKED and three indicator addresses. It lists no workaround; the jump-box limit and Threat IDs 510046 and 510047 are partial cover only. | [source](https://security.paloaltonetworks.com/CVE-2026-0322) | | 14 Oct 2026 | NVD publishes the record the same day. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-0322) | | 14 Oct 2026 | CISA adds the CVE to KEV with a due date of 17 Oct 2026. Ransomware use is listed as Unknown. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322) | | 15 Oct 2026 | Palo Alto records an update to the advisory. | [source](https://security.paloaltonetworks.com/CVE-2026-0322) | | 17 Oct 2026 | KEV due date, 3 days after disclosure and listing. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322) | | 5 Nov 2026 (SonicOS CVE-2026-19868) | SonicWall releases advisory SNWLID-2026-0014, version 1.0, which fixes seven CVEs. For CVE-2026-19868 (CVSS 3.1 7.1) a user holding valid credentials for LDAP, RADIUS or TACACS+ can bypass SAML authentication on the SSLVPN, which also defeats SAML-enforced MFA. It applies only where SAML is enabled for SSLVPN logins. Fixes: Gen7 7.3.4-7013 and Gen8 8.2.3-8015 or higher. | [source](https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014) | | 6 Nov 2026 (SonicOS CVE-2026-19868) | NVD publishes the record. CISA adds the CVE to KEV with a due date of 20 Nov 2026, and its SSVC entry records exploitation as active and automatable as no. SonicWall says its PSIRT investigated a report indicating potential exploitation; neither it nor CISA gives a first attack date or indicators. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19868) | | 9 Nov 2026 (SonicOS CVE-2026-19868) | SonicWall advisory version 1.1 updates the workaround: if SAML is in use and you cannot upgrade, disable or delete every active LDAP, RADIUS and TACACS+ authentication server on the firewall. | [source](https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014) | | 10 Nov 2026 | Siemens adds CVE-2026-0322 to advisory SSA-104023 for RUGGEDCOM APE1808 devices running PAN-OS Virtual NGFW and sends customers to Siemens support for patch information, 27 days after the Palo Alto advisory. | [source](https://cert-portal.siemens.com/productcert/html/ssa-104023.html) | Disclosure and KEV listing came on the same day, so there was no gap in which to wait: the exposure check in hour 1 and the jump-box rule by hour 4 would have been in place before the 17 Oct due date, three days later. The fixed build depended on the train, and embedded copies such as the Siemens appliances waited 27 days for a vendor path, so the inventory in hour 1 has to include third-party devices that run PAN-OS. Because the flaw needs a high-privilege account, the administrator list review in hour 4 matters as much as the version. ## Related pages - [Cisco ISE network access control playbook: the first 72 hours after a critical flaw](/playbooks/network-access-control/) - [Dependency playbook: the first 72 hours when a library you ship has a critical flaw](/playbooks/dependency/) - [Container base image playbook: the first 72 hours when a flaw is in the image your services build on](/playbooks/container-base-image/) ## Sources - Palo Alto Networks advisory CVE-2026-0322: https://security.paloaltonetworks.com/CVE-2026-0322 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-0322: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-0322 (read 19 Nov 2026) - NVD record for CVE-2026-0322: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-0322 (read 19 Nov 2026) - Siemens ProductCERT advisory SSA-104023: https://cert-portal.siemens.com/productcert/html/ssa-104023.html (read 19 Nov 2026) - SonicWall advisory SNWLID-2026-0014 (CVE-2026-19868): https://psirt.global.sonicwall.com/vuln-detail/SNWLID-2026-0014 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-19868: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-19868 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/network-edge-appliance/ --- # Cisco ISE network access control playbook: the first 72 hours after a critical flaw Playbook, sources read 19 Nov 2026 **The short answer.** Cisco Identity Services Engine (ISE) decides which users and devices join wired, wireless and VPN networks, so an authentication bypass there reaches everything behind it. Cisco rated CVE-2026-76460 at CVSS 10.0 on 16 Sep 2026, and CISA listed it the same day with forensic triage required. This playbook is for Cisco Identity Services Engine (ISE) and ISE Passive Identity Connector (ISE-PIC), the network access control (NAC) server that sets who joins the network. It is not a playbook for a cloud identity provider such as a single sign-on service. An ISE node holds the access policy, the connections to the directory, RADIUS shared secrets, certificates and API client credentials. Flaws in it are usually authentication bypasses on an API or admin endpoint, reached from the network with no credentials. CVE-2026-76460 is one: Cisco says a successful attack may give command execution as root, and control of the server is control of network access. Patching is hard because the server must keep answering. A NAC outage locks out staff, VPN users and wireless devices, so nodes are patched one at a time in a set order. Evidence is also weak: after a root compromise the attacker can hide or remove traces on the node, so logs held elsewhere carry the investigation. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | List every node of the system, the release and the patch level. On Cisco ISE and ISE-PIC run `show version` at the CLI of each node, because a node can lag its peers. Mark which nodes are primary admin, secondary admin, policy service and PIC.
The fixed patch differs per release (3.1 Patch 12, 3.2 Patch 11, 3.3 Patch 12, 3.4 Patch 7, 3.5 Patch 4). Release 3.0 has reached end of software maintenance and has no fix: Cisco says to migrate to a supported release that includes the fix, so plan 3.0 nodes as an upgrade to one of those trains and then its fixed patch, following the Upgrade Guides on Cisco's ISE support page. | Node, role, release, patch level, and whether it is below the fixed patch. | | 0 to 4 hours | Platform engineering | List which networks can open connections to each node's management and API interfaces, from firewall rules and from a test on a normal user VLAN and any guest network. Then apply infrastructure ACLs so only required management and control plane traffic reaches the nodes. List the admin stations and integrations that call ISE before you apply, since anything not listed loses access. Cisco's advisory does not name the API endpoint or port, only that the bypass reaches "the web-based management interface", so treat every management-plane port the node listens on as the target. Test from outside your network, for example from a cloud host or a phone hotspot: nmap -Pn -p 443,8443,9060 , and curl -k -sI --connect-timeout 5 https:///admin/. Every port must show filtered or time out.
Cisco has no workaround, and ACLs are its only mitigation; they reduce who can reach the API but do not fix the flaw. | The ACL before and after, the callers you kept, and the test from an untrusted VLAN. | | 0 to 4 hours | Security lead | Take a copy of the configuration and a backup before touching a node: Administration > System > Backup & Restore. Start a support bundle on each node with debug logs and shared key encryption, so the apigateway access log is kept.
Patching restarts services, and the log on a node may be the only record of what ran against the API. | Backup name, support bundle name, hash, time, and who holds the shared key. | | 4 to 24 hours | Security lead | Review the access log on every node of the deployment for suspicious usernames: `show logging application ise-kong/access.log \| include dummyuser`, where dummyuser is the example string in Cisco's advisory. Replace it with each username you want to look for: any name that is not an account you created, and any name an attacker could pick, since Cisco calls its example non-exhaustive. Read ./ise/logs/apigateway/access.log.gz from the support bundle for older entries. Any hit on a username nobody owns is a possible compromise.
Cisco names this log and this command as its indicator, and CISA marks the entry forensic triage Yes, so the search must be run per node, not on the primary only. | Per node: period covered, usernames searched, and each hit with the source address. | | 4 to 24 hours | Security lead | Cross-check network and firewall logs outside the device for uploads from the ISE node to external addresses and downloads from malicious addresses. Cisco publishes no indicator addresses, so look for any outbound flow from the node that has no purpose.
After root access the evidence on the box can be removed, so the network record is the independent check. | The query, the period, and each unexplained flow. | | 4 to 24 hours | Platform engineering | Install the fixed patch one node at a time, secondary admin and policy service nodes first, the primary admin last. Confirm replication and `show application status ise` after each node, and test a wired, wireless and VPN login before the next node.
Authentication keeps working only if a node is healthy while its peer restarts. | Per node: patch level, time, replication state, the three login tests. | | 4 to 24 hours | Platform engineering | Rebuild any node with a suspicious hit instead of patching it: re-image, then restore the configuration backup. Cisco strongly recommends re-imaging if you suspect compromise.
A patch closes the hole but does not remove persistence left by someone who already had root. | Which nodes were rebuilt, from which backup, and the date. | | 24 to 72 hours | Security lead | If a node may have been compromised, rotate everything it holds: admin passwords, the AD or LDAP join and bind accounts, RADIUS shared secrets, pxGrid and ERS API clients, certificates (Administration > System > Certificates), and any integration credentials. End active admin and portal sessions.
Whoever had root can read every secret the system uses to talk to other systems. | Each secret, its owner, the rotation date, and the dependent system tested. | | 24 to 72 hours | Application owner | Test sign-in for each application and network service that depends on the system, including guest and posture flows, and record failures. Review accounts that were created in the exposure window and any policy change you cannot explain.
Rotations and ACLs break integrations quietly, and a new policy or account is how a bypass turns into lasting access. | Test list with result, and the list of new accounts and policy changes with owner confirmation. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet If the vendor has no fix yet, the mitigation for a NAC server is to shrink who can reach it. Allow only the admin hosts and the devices that must talk to it, and put the API behind the management network. The cost is that every integration outside the list stops working, so build the caller list first. Cisco's guidance for ISE treats ACLs as a reduction of exposure, not a fix. Take the system offline only if exposure is wide and exploitation is reported, and only after you know what fails without it. An offline NAC server means wired and wireless authentication fails open or fails closed depending on your design, so decide that in advance. When compromise is suspected, a rebuild from a clean image and a configuration backup is the safe route, and the cost is node downtime. ## CVE-2026-76460 worked through, hour by hour Cisco published advisory cisco-sa-ISE-ABP-VNSW7Tn5 on 16 Sep 2026 at 16:00 GMT for an authentication bypass in an API endpoint of Cisco ISE and ISE-PIC. The CVSS 3.1 base score is 10.0 and no configuration is required to be affected. Each row pairs a published event with what a team running this playbook does at that hour. The advisory, the NVD record and the KEV entry all appeared on day 0, and no source dates the first attack. | When | What happened | Source | | --- | --- | --- | | 16 Sep 2026, 16:00 GMT (hour 0) | Published: Cisco releases version 1.0 with fixed patches for 3.1 to 3.5, says release 3.0 is end of software maintenance, says there is no workaround, names the ise-kong access.log search for suspicious usernames on every node as the indicator, gives no attack date, attacker detail or indicator addresses, and says its PSIRT is aware of active exploitation. The bug (CSCww39530) was found while resolving a TAC support case. Response: the security lead reads the advisory and starts the version check with show version on every node. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 16 Sep 2026, 17:00 to 20:00 GMT (hours 1 to 4) | Response, no new publication: the inventory is complete, with each node marked by release, patch level and whether it sits below the fixed patch. Platform engineering lists the callers of each node, applies the ACLs, which are the one mitigation the advisory gives, and runs the outside port test, while the security lead takes the configuration backup and starts the support bundle with debug logs so the apigateway access log is kept. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 16 Sep 2026, 21:17 UTC (hour 5) | Published: NVD publishes the record with a CVSS 3.1 base score of 10.0 (Critical), supplied by Cisco. Response: the per-node log search for usernames nobody owns is under way, with network and firewall logs checked for outbound flows from each node. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76460) | | 16 Sep 2026 (CISA publishes no time; hours 0 to 8) | Published: CISA adds the CVE to KEV with a due date of 19 Sep 2026, forensic triage Yes and ransomware use Unknown. Its SSVC entry in NVD records exploitation as active, automatable as yes and technical impact as total. Response: the forensic triage flag moves the log search ahead of the patch, so no node is patched before its logs and support bundle are copied. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | | 17 Sep 2026 (hours 8 to 32) | Response, no new publication: nodes with a hit are re-imaged and restored from the backup. The remaining nodes are patched one at a time, secondary admin and policy service nodes first, the primary admin last, with a wired, wireless and VPN login test after each. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 18 Sep 2026 (hours 32 to 56) | Response, no new publication: if any node had a hit, the AD or LDAP accounts, RADIUS shared secrets, pxGrid and ERS clients and certificates are rotated, and admin and portal sessions are ended. | [source](https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5) | | 19 Sep 2026 (hour 72) | Published: KEV due date, three days after listing. Response: every node is on its fixed patch or documented as an exception, the integration tests are recorded, and the new accounts and policy changes in the exposure window are confirmed by their owners. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460) | The advisory and the listing came on the same day, so the 72-hour window of the playbook ends at the KEV due date. Because KEV marks forensic triage Yes, a patched node still needs its log search and network cross-check before it is called clean. The advisory gives no attack date, so the log search must cover the whole period the nodes were reachable, not only the days after 16 Sep. ## Related pages - [Container base image playbook: the first 72 hours when a flaw is in the image your services build on](/playbooks/container-base-image/) - [The network edge appliance playbook for the first 72 hours](/playbooks/network-edge-appliance/) - [SaaS vendor playbook: the first 72 hours when the flaw is in a service you cannot patch](/playbooks/saas-vendor/) ## Sources - Cisco Security Advisory cisco-sa-ISE-ABP-VNSW7Tn5: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-ISE-ABP-VNSW7Tn5 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-76460: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-76460 (read 19 Nov 2026) - NVD record for CVE-2026-76460: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-76460 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/network-access-control/ --- # Container base image playbook: the first 72 hours when a flaw is in the image your services build on Playbook, sources read 19 Nov 2026 **The short answer.** A flaw in a container base image sits in every image built on it and in none of your repositories. This playbook gives the registry and cluster commands to find affected images, the order to rebuild them, and the policy that blocks the old tag, worked through the Alpine OpenSSH bypass of 5 Nov 2026. A base image makes one flaw into hundreds. Every service that starts from the same FROM line carries the vulnerable package, and so does every intermediate image another team built from it. The flaw is exploited wherever the package runs, so the first question is not 'which images contain it' but 'which images run it, and where'. In CVE-2026-123158 the package was sshd in Alpine's openssh-server. Most Alpine images never start sshd, and those are not exposed. An image that does start it, with password login on, lets an attacker log in without a password. Patching is the hard part, because the fix is not applied to anything that runs. A fixed package in the distribution repository changes nothing until an image is rebuilt, pushed under a new tag or digest, and the old containers are replaced. Pinned digests, cached build layers and images built from your own base images all keep the old package alive. The steps below follow that chain: find, narrow, mitigate, rebuild in order, replace, block, and check. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Platform engineering | List every image running in the cluster, with init containers, and keep the list as your scope. kubectl get pods -A -o jsonpath='{range .items[*]}{range .spec.containers[*]}{.image}{"\n"}{end}{range .spec.initContainers[*]}{.image}{"\n"}{end}{end}' \| sort -u > running-images.txt. For hosts that run plain Docker, docker ps --format '{{.Image}}' \| sort -u. Take the list from the running system, not from the registry: the registry also holds images nobody runs.
Running images set the work and the order; the registry gives a longer list that delays the first answer. | running-images.txt with the time, cluster and context name, and the image count. | | 0 to 4 hours | Platform engineering | Find which images contain the affected package, without running them. For Alpine the package database is at lib/apk/db/installed, so for one image: crane export "$img" - \| tar -xOf - lib/apk/db/installed \| grep -A1 -x 'P:openssh-server'. The line after P: is V:, the version. Loop over running-images.txt and over the registry: crane catalog registry.example.com, then crane ls . Fixed openssh-server versions per Alpine branch: edge 10.6_p2-r0; 3.24 10.3_p1-r2; 3.23 10.2_p1-r1; 3.22 10.0_p1-r11; 3.21 9.9_p2-r1; 3.20 9.7_p1-r6; 3.19 9.6_p1-r3; 3.18 9.3_p2-r4; 3.17 9.1_p1-r7; 3.16 9.0_p1-r6. Compare the package version, not the banner printed by ssh -V. A scanner gives the same answer: syft "$img" -o json \| jq -r '.artifacts[] \| select(.name=="openssh-server") \| .version'. Debian-based images use the same idea with /var/lib/dpkg/status.
Reading the package database is quick and safe on any image, including ones whose entrypoint would start a service. | A table of image, digest, package, version and whether it is in the fixed list; keep the script. | | 0 to 4 hours | Platform engineering | Find the images built on a vulnerable base even when their Dockerfile says nothing useful. Layer check: save the base and the candidate configs with crane config "$base" > base.json and crane config "$app" > app.json, then jq -n --slurpfile b base.json --slurpfile a app.json '($a[0].rootfs.diff_ids[:($b[0].rootfs.diff_ids\|length)]) == $b[0].rootfs.diff_ids' prints true when the candidate starts with every layer of the base. Source check: grep -rEn '^FROM .*alpine' --include='Dockerfile*' . across your repositories, and note every FROM that names one of your own images, since those are intermediates that also need a rebuild.
Intermediate images built from your own base hide the dependency from a search on the application repository. | A base-to-image tree with the digest at each node, saved with the date. | | 0 to 4 hours | Security lead | Cut the list to images that run the vulnerable daemon and are reachable. Does it start sshd: crane config "$img" \| jq '.config.Entrypoint, .config.Cmd, .config.ExposedPorts' and look for sshd or 22/tcp. Is the exposed condition true in a running container: kubectl exec -- sshd -T \| grep -iE 'passwordauthentication\|usepam\|permitrootlogin' (sshd -T needs host keys, so run it in the container started normally), then kubectl exec -- awk -F: '$2=="*"{print $1}' /etc/shadow for accounts with no password, and awk -F: '$2 ~ /^\$[56]\$/{print $1}' /etc/shadow for accounts with a SHA-256 or SHA-512 hash. Both lists non-empty with PasswordAuthentication yes and the port reachable puts the container in scope. Containers that do not run sshd are not affected, according to Alpine.
Alpine's own advisory excludes containers without sshd, so this step removes most of the list and focuses the rebuild on what is exploitable. | Per container: sshd yes or no, PasswordAuthentication value, UsePAM value, the account names found, port reachability. | | 4 to 24 hours | Platform engineering | For in-scope containers that cannot be replaced within hours, set PasswordAuthentication no. Ship it as a mounted sshd_config or a drop-in in /etc/ssh/sshd_config.d/, check it in the container with sshd -t, and confirm with sshd -T \| grep -i passwordauthentication after a restart of sshd. Look in Match blocks too. Before you roll it out, prove key login still works from a second session, because the change locks out every password user. MaxAuthTries 1 is a third-party stopgap that Medcrypt reports keeps password login while blocking its reproduction; treat it as a bridge only.
OpenSSH and Alpine both name this as the immediate step, and it closes the hole in minutes while the rebuild takes hours. | The config change, the sshd -T output after, who tested key login, and an end date for the exception. | | 4 to 24 hours | Platform engineering | Rebuild in layer order: base, then intermediates, then applications. In the base Dockerfile add RUN apk upgrade -U -a, which is the update command Alpine gives, and build with docker build --pull --no-cache so nothing comes from cache. Alpine point releases 3.21.10, 3.22.8, 3.23.8 and 3.24.4 contain the fixed package. Branches 3.16 to 3.20 got no new point release, so a tag like alpine:3.20 does not help and the upgrade line is required. Change any FROM alpine@sha256:... pin to the new digest. Verify each build: docker run --rm --entrypoint apk info -v openssh-server, and compare it with the fixed version for the branch in step 2.
Intermediates built from the old base keep the old package until they are rebuilt, so the order is what makes the dependent rebuilds correct. | The new digest of each image, the package version printed by apk info, the CI run, and the build time. | | 4 to 24 hours | Application owner | Run the acceptance tests on the rebuilt image before rollout. Key login must work: ssh -o PreferredAuthentications=publickey user@host true. Password login must be refused on images that keep PasswordAuthentication no: ssh -o PubkeyAuthentication=no -o PreferredAuthentications=password user@host true must fail. Run the service's own health checks, and check that anything that logged in over SSH with a password, such as a deploy script or a backup job, has been moved to keys.
The owner knows the jobs that quietly depend on password SSH, and a rollout that breaks them ends in a rollback and a new gap. | Test names, results, the jobs found that used password login, and the approval with the digest it covers. | | 4 to 24 hours | Platform engineering | Preserve evidence, then replace. For each in-scope container reachable from the internet, save the filesystem before it is deleted: docker export > ctr-.tar, or kubectl cp :/var/log ./evidence and the same for /etc. Roll out the new image: kubectl set image deployment/ = and kubectl rollout status deployment/. Then prove nothing runs the old image by digest, not tag: kubectl get pods -A -o jsonpath='{range .items[*]}{.metadata.namespace}/{.metadata.name} {range .status.containerStatuses[*]}{.imageID} {end}{"\n"}{end}' \| grep -F . An empty result closes the step.
A pod replaced before it is examined takes its logs with it, and imageID is the only field that shows the digest actually running. | The archived tar with its hash, the rollout output, and the empty grep result with the time. | | 24 to 72 hours | Platform engineering | Block the old tag and digest so no deploy can bring them back. Admission policy, Kubernetes 1.30 or later, as a ValidatingAdmissionPolicy with validation expression !object.spec.containers.exists(c, c.image.contains('myapp:1.4.2')) and the message 'image has CVE-2026-123158', with matchConstraints on pods, then a ValidatingAdmissionPolicyBinding with validationActions: [Deny]. Match the digest as well as the tag. In the registry remove or retag the old tag, for example aws ecr batch-delete-image --repository-name myapp --image-ids imageTag=1.4.2, or set the repository to immutable tags and publish the fixed image under a new tag. Cost of deleting: you lose rollback to that image, so keep a tested fixed rollback target first.
Without a block, a rollback, a stale Helm value or a cached Dockerfile redeploys the old package weeks later. | The policy name and manifest, the registry action and who ran it, and a failed test deploy of the old tag. | | 24 to 72 hours | Security lead | Hunt for logins that already happened, on every in-scope container that was exposed. Container logs: kubectl logs \| grep 'Accepted password for', and look for accounts that have no password. Check the archived filesystems and live containers for the traces GreyNoise reports, such as /etc/init.d/tac-agent and a tac[.]vpnla[.]org entry in /etc/hosts. If you cannot rule out a login, Alpine says to treat the machine as compromised: rotate every secret mounted in the container and review the cloud roles it held.
Hunting waits until the fix is rolled out so that the window closes first, but the first intrusions came within two days of the release. | The logs searched with time ranges, files checked, and a list of containers declared clean or compromised, with the evidence for each. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet No fixed image can mean three things. The distribution has a package but no new point release (Alpine 3.16 to 3.20), so the fix reaches you through apk upgrade in the build and not through the tag. The image is on a release that no longer gets updates, or its vendor ships no fix. Or an appliance image bundles the package and you cannot rebuild. Your options are in order of cost. Set PasswordAuthentication no. Remove the daemon the image does not need: containers rarely need sshd, kubectl exec replaces it, and apk del openssh-server makes the flaw unreachable. Build from the upstream fix: Medcrypt reports the OpenSSH fix commit applies cleanly to OpenSSH 7.3p1 and later, at the cost of a package you now maintain. Take the container offline, or block port 22 at the network policy or security group, when all three are true: sshd is reachable from untrusted networks, an account has no password while another has a SHA-256 or SHA-512 hash, and the password setting cannot change. OpenSSH's own wording for musl systems without PAM is to set PasswordAuthentication=no or disable SSH immediately. Disabling SSH cuts remote administration of that image, so make sure the cluster or console access works first. ## CVE-2026-123158 worked through, hour by hour The OpenSSH password bypass of 5 Nov 2026 is a base-image case because the affected package was Alpine's openssh-server, built without PAM against musl libc. The CVSS base score of 8.1 comes from the MITRE CVE record; NVD has no rating of its own. This table keeps the moments that matter for an image. The full dated sequence, from the fix commit to the exploit code, is in the [Linux server SSH playbook](/playbooks/linux-server-ssh/). | When | What happened | Source | | --- | --- | --- | | 5 Nov 2026, 09:00 UTC (hour 0) | Published: the OpenSSH fix is public at 09:00 and 10.6p2 is released as an urgent security release at 09:01. Alpine publishes its advisory and fixed packages the same day, including the 3.21.10, 3.22.8, 3.23.8 and 3.24.4 point releases. Response: platform engineering writes running-images.txt from the cluster (step 1). | [source](https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/) | | 5 Nov 2026, hours 0 to 4 | Response, no new publication: the apk database is read from each running image and compared with the fixed versions, base and intermediate images are traced by layer, and the sshd and PasswordAuthentication check cuts the list to containers that run the daemon (steps 2 to 4). Alpine says containers that do not run sshd are not affected. | [source](https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/) | | 5 Nov to 6 Nov 2026, hours 4 to 24 | Response, no new publication: PasswordAuthentication no goes in where replacement takes longer than hours, the base is rebuilt with apk upgrade -U -a and --no-cache, then intermediates and applications. Exposed containers are exported, replaced by digest, and the imageID grep for the old digest returns nothing (steps 5 to 8). | [source](https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/) | | 7 Nov 2026, 00:46 UTC (about hour 40) | Published: GreyNoise observes the first successful exploitation, on one of its own sensors. Intrusions with follow-up commands begin the same day. Response: the exposed containers were replaced about 16 hours earlier. | [source](https://sshole.events.greynoise.io/) | | 8 Nov 2026, 09:00 UTC (hour 72) | Response, no new publication: the admission policy and registry block for the old tag and digest are in place, and the hunt for accepted passwords and the GreyNoise traces is under way (steps 9 and 10). | [source](https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/) | | 9 Nov 2026 | Published: CISA adds the CVE to the KEV catalog, due 12 Nov 2026. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | The 3.16 to 3.20 branches are the case where the tag does not move: only the apk upgrade line in the build and the imageID check at step 8 show the fix landed. A team that started at the KEV entry on 9 Nov began 2 days after the first success on a GreyNoise sensor. ## Related pages - [SaaS vendor playbook: the first 72 hours when the flaw is in a service you cannot patch](/playbooks/saas-vendor/) - [Cisco ISE network access control playbook: the first 72 hours after a critical flaw](/playbooks/network-access-control/) - [Email gateway playbook: the first 72 hours after a critical flaw](/playbooks/email-gateway/) ## Sources - Alpine Linux security advisory, 5 Nov 2026: https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/ (read 19 Nov 2026) - OpenSSH 10.6p2 release notes: https://www.openssh.org/releasenotes.html#10.6p2 (read 19 Nov 2026) - OpenSSH fix commit: https://api.github.com/repos/openssh/openssh-portable/commits/4353fdc43ef646c43a8dda7025cc1a898caf1482 (read 19 Nov 2026) - CISA KEV feed, catalog version 2026.11.18: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026) - GreyNoise event page and timeline: https://sshole.events.greynoise.io/ (read 19 Nov 2026) - Researcher write-up by Ozgun Kultekin: https://sshhole.com (read 19 Nov 2026) - Medcrypt reproduction and mitigation guidance: https://www.medcrypt.com/blogs/asterisk-hole-cve-2026-123158-when-a-locked-account-is-the-way-in (read 19 Nov 2026) - NVD record for CVE-2026-123158: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/container-base-image/ --- # SaaS vendor playbook: the first 72 hours when the flaw is in a service you cannot patch Playbook, sources read 19 Nov 2026 **The short answer.** When the flaw sits in a hosted product, nothing is yours to patch, so the work is proving whether your tenant was hit. This playbook gives 27 questions to send the vendor, one per line in six timed batches, plus the own-side steps, worked through the BeyondTrust BT26-05 case day by day. In a SaaS product the vendor owns the code, the servers and the logs. Flaws are exploited against the vendor's shared platform, and the attacker gains whatever the platform holds for you: stored credentials, API keys, SSO configuration, customer data, and often a path back into your network through the integrations you set up. Nothing to patch does not mean nothing to do. The vendor often patches before you hear of the flaw: BeyondTrust patched its hosted Remote Support instances by 9 Oct 2026, six days before the 15 Oct disclosure. What is hard is that your questions depend on the vendor's answers, and the vendor holds the only logs. You cannot see whether your tenant was in the exposure window, only ask. This page therefore gives the complete question list, the order in which to send it, and the steps you can take without waiting for any reply. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Read the vendor's advisory, status page and any CISA KEV entry. Sort your product into one of three cases: hosted by the vendor and already patched, hosted and not yet patched, or self-hosted (customer-managed) so that you do patch. For BeyondTrust the advisory split exactly this way: as of 9 Oct 2026 the patch was applied to all Remote Support SaaS customers, while self-hosted customers had to apply BT26-05 themselves. Note the date the vendor says it learned of exploitation and the first exploitation date it gives.
The case decides the whole plan: a patched tenant still needs the exposure-window questions, an unpatched one needs controls, and a self-hosted one belongs to the asset-type playbook. | Product, case (patched, unpatched, self-hosted), source link and the time you read it. | | 0 to 1 hour | Security lead | List what the vendor holds or can reach for you: data categories, API keys and tokens you issued to it, SSO and SAML connections, stored or injected credentials, remote-access paths into your network, webhooks, and the vendor's agents on your machines. Mark each item with the person who owns it.
You cannot ask 'what did the attacker reach' until you know what there was to reach. | The list, one line per item, with owner and where the secret lives on your side. | | 1 to 4 hours | Security lead | Send the status batch in writing to the vendor's security contact and your account manager, and ask for a reply time.
Q1. Is the product and edition we use affected by the CVE or advisory named in your notice? Quote its identifier in the message. Name the components and the region our tenant runs in.
Q2. Is our tenant patched? Give the date and UTC time the fix reached it, and the build or version string it now reports.
Q3. From what date was our tenant running the vulnerable code, and was it reachable from the internet in that period?
Q4. Is any part of the fix left to us (agents, connectors, plug-ins, settings, on-premises components we installed)? Name each and the version we need.
Q5. Which compensating controls do you advise until our tenant is fixed, and what does each one break?
Q6. Who is our named incident contact, how do we reach them out of hours, and when is your next update?
These six answers set the exposure window and tell you whether any part of the fix is yours to do. | Date and time sent, who it went to, the ticket number, and each answer with the time it arrived. | | 1 to 4 hours | Platform engineering | Reduce what the vendor can reach, without rotating yet. Restrict the vendor's inbound access to your network to the addresses it publishes; require SSO with phishing-resistant MFA for admin sign-in; disable dormant integrations and API accounts (disable rather than delete, so the audit trail survives); lower session lifetimes. Export your own side logs now (IdP sign-ins, API gateway logs, firewall logs for the vendor's addresses). Hold rotation of the secrets the vendor stores until it confirms the fix: Mandiant's BeyondTrust guidance says to rotate credentials only after the platform is patched, because new credentials are exposed to an attacker who still has access.
Controls you own can be changed in minutes; secrets held by a vendor that is still exposed cannot be safely replaced yet. | Each control changed, time, and the log export file names. | | 4 to 24 hours | Security lead | Send the compromise batch.
Q7. Has the flaw been exploited against any customer? Give the date of the first known exploitation, how it was found and by whom.
Q8. Was our tenant accessed by anyone other than us during the exposure window? Answer yes, no or unknown, and give the evidence behind the answer.
Q9. If the answer is unknown, what evidence is missing: which logs, how long you keep them, and whether the attacker could have deleted them?
Q10. Which indicators can you share (addresses, hashes, user agents, file names, account names), and which behaviors should we hunt for if the actor changes infrastructure per victim?
Q11. Did any of your monitoring alert on the activity, and how was each alert handled (human review, automatic closure)?
Q12. Which of your systems did the attacker reach (shared infrastructure, other tenants, support tools, build systems, corporate network), and could our data be read from any of them?
These questions test whether your tenant was reached, and what the vendor's own detection saw; BeyondTrust's sensors fired on the privilege escalation and its automated triage closed the alerts as false positives, which only a question like Q11 brings out. | Each answer in the incident record, with 'unknown' kept as unknown and the date it was given. | | 4 to 24 hours | Security lead | Send the data and credential batch.
Q13. What data of ours do you hold, by category and location, and which of it was readable from the affected component?
Q14. Which secrets of ours do you store (API keys, OAuth tokens, SSO signing keys, vault or integration credentials), which were readable, and in what order do you advise we rotate them?
Q15. Which of those secrets can we rotate ourselves, which need you, and how long do you need for the ones you hold?
Q16. Were sessions, tokens and API keys issued by your platform invalidated when the fix went in? If not, will you invalidate them for our tenant?
Q17. Did the attacker create accounts, roles, API keys, webhooks, integrations, outbound destinations or MFA registrations in our tenant? List each with its creation time. Ask the vendor to put each answer in writing even when the answer is 'none'.
Your rotation order and your notice duties both depend on what was readable. | Per data category and secret: readable yes, no or unknown, and your rotation owner. | | 4 to 24 hours | Security lead | Send the evidence batch.
Q18. Can you export our tenant's audit, access, admin and API logs in raw form, from 90 days before the earliest exploitation date to today, and by what date?
Q19. What is the retention period for those logs, are they stored off the application hosts, and do they show gaps or deletions in the window?
Q20. Will you place a legal hold on our tenant's logs, snapshots and any forensic images that include our data? For hosted tenants ask the vendor's support to start forensic preservation, as BeyondTrust's guidance does for SaaS tenants.
Logs age out, and the actor in the BeyondTrust case deleted logs on the systems it reached. | Export request id, the date range asked for, the date promised, the hold confirmation. | | 4 to 24 hours | Security lead | Hunt in your own logs for the vendor's indicators and for the actions an intruder in your tenant would take. In the tenant's audit log (once exported): new users, role changes that raise someone to administrator, new API keys, new MFA registrations, new webhooks or outbound destinations, and sign-ins from commercial VPN or residential addresses you cannot tie to a person or a ticket. In your IdP and firewall: sign-ins and connections from the vendor's service accounts at unusual hours, and any use of the keys you issued to the vendor. Go back to the first exploitation date the vendor gives, not to disclosure day.
The vendor's answers may be slow or incomplete, and your own logs are the evidence you can check yourself. | The queries run, the time range, hits and non-hits. | | 24 to 72 hours | Security lead | Send the closing batch and set a reply date for every open question.
Q21. What was the root cause (weakness class and component), and did variant analysis find other paths to it?
Q22. What evidence shows the fix works: retest of every supported version, an independent test, or both?
Q23. What are you changing in detection and response so the same activity is caught earlier?
Q24. Will you give us a written incident report with a date, and tell us when new facts change your answers? Escalate unanswered items through your account executive and the contract's security contact on the date you set. Rotate the secrets from step 2 once the vendor confirms the fix, in dependency order: administrator credentials first, then users, API secrets, SSO signing keys and bind accounts, stored vault or integration credentials rotated at the target system, then anything typed into a session during the window.
By hour 24 the vendor has had time to answer the facts, and root cause plus fix evidence tell you whether to trust the fix. | Answers with dates, the list of open questions with owner and next date, and the rotation record. | | 24 to 72 hours | Executive | Take the contract position.
Q25. When did you first learn of the flaw, of exploitation, and of exploitation in our tenant, and when did you tell us? (Compare with the notice period in our contract.)
Q26. Will you notify regulators or affected people on our behalf, or do you expect us to? What written statement can you give us for our own notices?
Q27. What remedies apply under the contract: service credits, reimbursement of forensic costs, audit rights and a right to end the agreement early? Read the agreement's incident-notice clause, service levels, audit rights and exit terms, and decide whether the vendor's notice time and its answers meet what you agreed. Decide whether any integration should be reduced to the least reach the work needs.
Remedies and exit rights are used in negotiation after the facts are known, and the vendor's own timeline is the baseline. | Notice time promised and delivered, remedies requested, decision and date. | | 24 to 72 hours | Communications | If the vendor says customer data was exposed or cannot rule it out, prepare your own notices. Your clock may start when you become aware, not when the vendor tells you: GDPR Article 33, for example, gives a controller 72 hours after becoming aware of a personal data breach to notify the supervisory authority. Take the vendor's written statement (Q26), ask counsel which duties apply, and prepare a customer note that says what is known and when you will update.
A legal clock can already be running on the day the vendor's email arrives. | The time you became aware, who decided, drafts and the notices sent. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet When the vendor has not yet fixed a hosted product, you have a few controls: restrict who can reach the tenant (IP allow lists, SSO only), cut or disable integrations, move the most sensitive data out of the product, and keep logs off the platform. Each costs access or convenience for users, and none replaces the fix. Do not rotate secrets the vendor stores while it is still exposed, since the new secrets reach the attacker too. Suspend use of the service when the vendor cannot say whether your tenant was reached and the product holds credentials that open your network (a remote-access tool, a password vault, an identity service), or when exploitation is confirmed and no fix date is given. CISA's KEV action text for such entries says to follow vendor instructions, apply BOD 26-04 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. The cost of suspending is a business process you run by hand for some days; the cost of continuing may be a path into your whole estate. ## BeyondTrust CVE-2026-102764 (BT26-05) worked through, day by day BeyondTrust Remote Support carried an origin-validation flaw that a suspected China-nexus actor used as a zero-day against one BeyondTrust-hosted instance and one self-hosted appliance. The hosted customer could patch nothing: the vendor found and fixed it. The sources give dates; hours are given only for the 27-minute containment. | When | What happened | Source | | --- | --- | --- | | 29 Aug 2026 | The actor exploits the flaw against one BeyondTrust-hosted Remote Support instance and one self-hosted appliance. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 27 Sep 2026 | A zero-day campaign against another vendor's edge appliances becomes public, and Mandiant sends BeyondTrust a non-specific note. Hunts start across its SaaS fleet. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 28 Sep 2026 | The hunt surfaces the affected SaaS instance. BeyondTrust contains it 27 minutes after the compromise is identified, with evidence preserved. About seven hours later Product Security reproduces the entry in a lab. Its sensors had fired when the actor escalated privileges, but an automated triage tool closed those alerts as false positives. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 30 Sep 2026 | Mandiant shares findings from the self-hosted customer, confirming BeyondTrust's own. The fix is validated on a test build. | [source](https://www.beyondtrust.com/blog/entry/bt26-05) | | 8 Oct 2026 | BeyondTrust ships the BT26-05 patch. The advisory says the patch has been applied to all Remote Support SaaS customers as of 9 Oct 2026. | [source](https://www.beyondtrust.com/trust-center/security-advisories/bt26-05) | | 15 Oct 2026 | Coordinated disclosure: advisory BT26-05, CVE-2026-102764 (CVSS v4 9.3), BeyondTrust's blog and Mandiant's analysis. BeyondTrust knows of two affected customers and has seen no activity since August. Mandiant says customers who use the hosted service should still follow its credential and configuration steps. | [source](https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust) | | 16 Oct 2026 | CISA adds the CVE to KEV, forensic triage required (Yes), due date 19 Oct 2026. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-102764) | A hosted customer first heard of the flaw on 15 Oct, 47 days after the 29 Aug exploitation and 6 days after the platform was patched, so no patch was left to apply and the useful work was entirely in the questions. Under this playbook the status batch (Q1 to Q6) would have gone out in the first four hours of 15 Oct and asked for the exposure window from 29 Aug; the compromise batch (Q7 to Q12) would have got the answer that one hosted instance was hit; the evidence batch (Q18 to Q20) would have secured logs the actor was known to delete on the systems it reached; and credential rotation, held until the vendor's patch was confirmed, would have followed the order Mandiant publishes. BeyondTrust itself needed 30 days from first exploitation to find the hosted intrusion (29 Aug to 28 Sep), and its fix took 10 more days (28 Sep to 8 Oct), which is why a customer should plan on the vendor's logs, not its alerts. ## Related pages - [Email gateway playbook: the first 72 hours after a critical flaw](/playbooks/email-gateway/) - [Container base image playbook: the first 72 hours when a flaw is in the image your services build on](/playbooks/container-base-image/) - [The remote support tool playbook for the first 72 hours](/playbooks/remote-support-tool/) ## Sources - BeyondTrust: ARMORY CASTLE and BT26-05: https://www.beyondtrust.com/blog/entry/bt26-05 (read 19 Nov 2026) - BeyondTrust security advisory BT26-05: https://www.beyondtrust.com/trust-center/security-advisories/bt26-05 (read 19 Nov 2026) - Mandiant and GTIG analysis of ARMORY CASTLE: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust (read 19 Nov 2026) - CISA KEV entry for CVE-2026-102764: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-102764 (read 19 Nov 2026) - NVD record for CVE-2026-102764: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-102764 (read 19 Nov 2026) - GDPR Article 33, notification of a personal data breach: https://gdpr-info.eu/art-33-gdpr/ (read 19 Nov 2026) - NIST SP 800-61 Rev. 3: https://csrc.nist.gov/pubs/sp/800/61/r3/final (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/saas-vendor/ --- # Email gateway playbook: the first 72 hours after a critical flaw Playbook, sources read 19 Nov 2026 **The short answer.** A mail gateway reads every message and faces the internet, so a flaw in it exposes mail content as well as the box. Using Fortinet's FortiMail CVE-2026-104286, this playbook gives the hour-by-hour order: check IBE, close webmail, preserve logs, hunt the published lines and addresses, then upgrade. A mail gateway is the one server that sees every inbound and outbound message in clear text, and its web interface (webmail, the secure-message portal, the admin console) is usually reachable from the internet so that outside recipients can open encrypted mail. Flaws in these interfaces are exploited by sending crafted web requests to that interface. CVE-2026-104286 is listed under path traversal (CWE-22) and improper neutralization of a NULL byte (CWE-158): an unauthenticated visitor sends crafted HTTP or HTTPS requests and writes arbitrary files on the appliance, and Fortinet lists the impact as execution of unauthorized code or commands. What the attacker gains is mail, not only a foothold. The indicators Fortinet published from compromised appliances show a root cron job, an admin session that logged out for no reason, and a new mail archive account whose destination is a remote server. An archive account like that copies mail passing through the appliance to a machine the attacker owns. What makes patching hard is that an upgrade means a reboot and a gap in mail flow, the workaround that works (turning off the IBE service) breaks encrypted mail delivery for some users, and older trains may have no fixed build at all. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | List every FortiMail appliance and virtual machine, including lab and disaster-recovery copies, and read its build. In the CLI run get system status and read the Version line, or open Dashboard > Status > System Information. Affected builds: 8.0.0 to 8.0.1, 7.6.0 to 7.6.6, 7.4.0 to 7.4.8, 7.2.0 to 7.2.9, and every 7.0 build. First fixed builds: 8.0.2, 7.6.7, 7.4.9; 7.2.10 was listed only as upcoming in the advisory revision of 7 Oct 2026; 7.0 has no fixed build and Fortinet says to migrate. FortiMail Cloud needs no action: Fortinet updated it on 5 Oct 2026 to 7.6.7 GA or 8.0.2 GA, and get system status on a Cloud instance should report one of those.
The version decides which of the later steps apply, and a 7.2 or 7.0 appliance can never be fixed by upgrading today. | One row per appliance: name, train, exact build, HA role, internet-facing yes or no. | | 0 to 1 hour | Platform engineering | Read the two settings that make the flaw reachable. IBE: GUI Encryption > IBE > IBE Service, or CLI show full-configuration system encryption ibe and look for set status enable. Web exposure: CLI show system interface and read set allowaccess on each interface for http and https, then from a host outside your network run nc -vz 443 and nc -vz 80. Also read the firewall and NAT rules that publish the appliance.
Fortinet ties its workaround and its log lines to IBE, but does not state IBE as a formal prerequisite, so an appliance with IBE off and a public web interface is still handled as exposed. | For each appliance: IBE status, interfaces that allow http or https, result of the outside port test. | | 1 to 2 hours | Security lead | Before changing anything, take a configuration backup and export Log > Event > System and Log > Encryption for the longest period the appliance retains, and copy them to storage the appliance cannot write to. Snapshot a virtual appliance. CISA's KEV entry marks forensic triage as required, and the actor's first step after exploitation is to alter the box.
Logs on the appliance are the evidence, and the workaround, the upgrade and a reboot all change them. | File names, hashes and the time the export finished; who held the copy. | | 2 to 4 hours | Platform engineering | Apply one workaround on each exposed appliance, in this order of preference. (1) Turn the IBE service off: GUI Encryption > IBE > IBE Service off, or CLI config system encryption ibe, set status disable, end. (2) If IBE must stay on, remove internet access to webmail and allow it only from trusted private networks, by firewall rule or by removing http and https from allowaccess on the public interface. (3) If a web application firewall sits in front, block POST requests to /ibe that contain ../ , the pattern Fortinet names. Re-run the read from step 2 afterwards.
Fortinet published the workaround at disclosure because exploitation was already reported, and it closes the path in minutes while a build takes a maintenance window. | Which option, time applied, who approved the loss of IBE delivery, and the result of the second port and setting read. | | 2 to 4 hours | Security lead | Search the exported logs for the lines Fortinet published. System event log: type=event subtype=system pri=debug user=system ui=cron msg="(root) CMD (/bin/sh -c 'O=/migadmin ..."; type=kevent subtype=admin pri=information user=admin ui=(null) action=logout status=success reason=unknown msg="User admin logged out from (null)."; and type=kevent subtype=config ... msg="Added 'archive234' to 'archive account'" with rotation-size[50], rotation-time[1], rotation-hour[14], destination[remote], remote-ip[79[.]141[.]169[.]187], remote-username[archive234], remote-directory[/uploads] (user: admin, from: cli). Encryption log: FortiMail::IBE::DecrypterMediaIn ... Caught BufferException(2), BufferImpl.cpp:973, 'Invalid Base64 Encoding at pos 0. Character=0x2a', followed by Internal user *@domain.tld failed to log in. Then open the archive account settings and look for any account you did not create, not only one named archive234.
A hit changes the job from patching to incident response before you upgrade and reboot. | Each matching line with its timestamp, and the full list of archive accounts with creator and date. | | 2 to 4 hours | Network security | Search firewall, proxy and NetFlow records for the appliance talking to 79[.]141[.]169[.]187 or 45[.]129[.]0[.]192, in both directions, for as far back as you keep records. Check specifically for outbound sessions from the appliance's own address, because the archive account sends mail to a remote server.
Fortinet gives exactly two attacker addresses, so a hit is strong evidence, and outbound traffic from the gateway is the sign mail was copied out. | Query text, time range searched, hits with first and last seen. | | 4 to 24 hours | Platform engineering | Upgrade to the first fixed build for the train: 8.0.2, 7.6.7 or 7.4.9, or later. Back up the configuration first. On an HA pair, upgrade the secondary, fail over, check mail flow, then upgrade the former primary. Warn the mail team that each node reboots. For 7.2, stay on the workaround until Fortinet lists 7.2.10 as released. For 7.0, start the migration to a fixed train now. After the reboot run get system status and confirm the new Version line.
The workaround stays in place until the version read proves the fix, because the upgrade is the riskiest change for mail flow. | Build before and after, time of each node, mail queue length before and after, and any bounced mail. | | 4 to 24 hours | Security lead | If any indicator matched, treat the appliance as compromised with root access: do not clean it in place. Rebuild from a clean fixed image and restore a configuration backup dated before the first indicator, then change every credential the appliance holds (admin accounts, directory bind accounts, TLS private keys and API tokens). Work out what was copied: the archive account's creation time marks the start, and its rotation settings show how often it sent mail out. Ask legal whether those messages include regulated data.
Fortinet's own indicators include a root cron entry, so file-level cleanup cannot be trusted. | Time of the first indicator, time the archive account was removed, mailboxes and domains in the copy window. | | 24 to 72 hours | Application owner | Test inbound and outbound delivery, spam filtering, DKIM signing and TLS policy, then re-enable IBE only on a build the version read confirms as fixed. Tell the users of IBE secure mail when delivery returns. Re-run the whole check list from step 1 on every appliance as a closing read, and keep the outside port test on the schedule.
A feature turned off as a workaround is easy to forget, and IBE users have no other path to open secure mail. | Time IBE came back, test messages sent, and the closing version table. | | 24 to 72 hours | Executive | Decide the open items with a date each: the 7.2 appliances waiting for 7.2.10, the 7.0 appliances that must migrate, and whether the web interface needs to face the internet at all. CISA gave 4 Oct 2026 as the due date for this entry, three days after listing, so the same 72 hours were the public deadline.
These are budget and mail-service decisions, not security-team decisions. | Owner and date for each open appliance. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet With FortiMail the fix gap is real: 7.2 had no released fix in the last advisory revision read, and 7.0 has none. The mitigations are the three in the workaround. Turning IBE off costs the secure portal: recipients can no longer open or reply to IBE messages, and users whose policies depend on IBE lose that delivery path. Closing webmail to the internet costs the same users, plus any outside user of webmail. The WAF rule costs writing and testing the pattern, and it only covers the traversal string that Fortinet named, so treat it as a second layer and not a reason to leave IBE on. Take the appliance offline when logs show an indicator, or when none of the three mitigations can be applied (for example the secure portal is needed and no WAF exists). CISA's action for the entry says to follow vendor instructions, apply BOD 26-04 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. Offline is less costly than it sounds: sending servers retry for days (RFC 5321 says to keep trying for at least 4 to 5 days), so a gateway outage of a day delays mail but need not lose it, provided a secondary route does not bypass the filtering you are trying to protect. ## CVE-2026-104286 worked through, day by day Fortinet's own Product Security team found the FortiMail flaw, and the advisory FG-IR-26-175 of 1 Oct 2026 already said it was reported as exploited in the wild. The sources give dates but no hours, so the table runs in days from disclosure. The first row is Day 0. | When | What happened | Source | | --- | --- | --- | | 1 Oct 2026 (Day 0) | Fortinet publishes FG-IR-26-175: CVSS 3.1 base 9.8 (temporal 8.8), Known Exploited: Yes, with the IBE workaround, two attacker addresses and the log lines. | [source](https://fortiguard.fortinet.com/psirt/FG-IR-26-175) | | 1 Oct 2026 (Day 0) | CISA adds the CVE to KEV the same day, forensic triage required: Yes, due date 4 Oct 2026, ransomware use Unknown. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286) | | 2 Oct 2026 (Day 1) | CISA's SSVC entry on the NVD record is stamped: exploitation active, automatable yes, technical impact total. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-104286) | | 4 Oct 2026 (Day 3) | KEV due date. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286) | | 5 Oct 2026 (Day 4) | Fortinet revises the advisory (solution update) and says it updated FortiMail Cloud to 7.6.7 GA or 8.0.2 GA that day. | [source](https://fortiguard.fortinet.com/psirt/FG-IR-26-175) | | 7 Oct 2026 (Day 6) | Second revision (FortiMail Cloud fix clarification). The table still lists 7.2 as 'Upgrade to upcoming 7.2.10 or above' and 7.0 as 'Migrate to a fixed release'. | [source](https://fortiguard.fortinet.com/psirt/cvrf/FG-IR-26-175) | An owner who started the clock at 1 Oct had the workaround, the IP list and the log lines in the first advisory, so steps 1 to 6 (version read, evidence copy, workaround, log hunt, address search) fit inside four hours of Day 0, and no fixed build was needed to close the hole. Upgrading 8.0, 7.6 and 7.4 appliances inside 24 hours would be done by 2 Oct, two days before the KEV due date of 4 Oct. The 72 hours also decided the 7.2 and 7.0 appliances: on 7 Oct, Day 6, the vendor table still had no released fix for them, so their only protection for at least those 6 days was the workaround, which is why the playbook puts the no-fix decisions in the 24 to 72 hour window and not later. No source gives a first exploitation date or a victim count, so the log hunt over the longest period the appliance retains is the only way to know whether a given appliance was hit before Day 0. ## Related pages - [The remote support tool playbook for the first 72 hours](/playbooks/remote-support-tool/) - [SaaS vendor playbook: the first 72 hours when the flaw is in a service you cannot patch](/playbooks/saas-vendor/) - [The artifact repository and CI system playbook for the first 72 hours](/playbooks/artifact-repository/) ## Sources - Fortinet PSIRT advisory FG-IR-26-175: https://fortiguard.fortinet.com/psirt/FG-IR-26-175 (read 19 Nov 2026) - Fortinet advisory FG-IR-26-175, CVRF revision history: https://fortiguard.fortinet.com/psirt/cvrf/FG-IR-26-175 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-104286: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-104286 (read 19 Nov 2026) - NVD record for CVE-2026-104286: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-104286 (read 19 Nov 2026) - FortiMail CLI reference, system encryption ibe: https://docs.fortinet.com/document/fortimail/7.6.0/cli-reference/6402/system-encryption-ibe (read 19 Nov 2026) - RFC 5321, section 4.5.4.1, sending strategy: https://www.rfc-editor.org/rfc/rfc5321#section-4.5.4.1 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/email-gateway/ --- # The remote support tool playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Remote support tools hold standing, trusted access to many endpoints. ConnectWise released ScreenConnect 26.6.5 on 8 Sep 2026 for CVE-2026-84869, and CISA listed it on 11 Sep. This playbook covers client versions, the file-transfer role setting, audit-log hunting and technician credential reset. A remote support tool is a legitimate way to run commands on endpoints, so an attacker who controls it needs no malware to move. Flaws in it usually let a low-privilege session or a rogue client move files or run code on the machine at the other end. In the ScreenConnect case (CVE-2026-84869) every version before 26.6.5 is affected, Cloud and On-Premise, in Support and Access sessions. ConnectWise's bulletin says the condition is in the ScreenConnect client and that "ScreenConnect servers are not impacted", yet its remediation for on-premises partners is still to upgrade the installation to 26.6.5 and then reinstall host clients and update access agents. The bulletin does not explain the link. Our reading, not the vendor's: the upgraded instance is where the 26.6.5 clients come from, so the server upgrade is the step that makes fixed clients available, and a console that shows 26.6.5 tells you nothing about the clients already installed. Patching has two parts. The server or cloud instance is upgraded, then every host client and access agent must be reinstalled or updated, and ConnectWise says not to rely on the console version alone. Detection is also hard: the activity looks like a technician doing normal work, so you hunt for actions no technician made. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Inventory every ScreenConnect instance (cloud and on-premises, including those bundled with Automate), the console version at Administration > Overview, and the version of each technician client and endpoint client. Read client versions across the fleet instead of on one workstation. PowerShell on a Windows host: `Get-ItemProperty HKLM:\Software\Microsoft\Windows\CurrentVersion\Uninstall\*,HKLM:\Software\WOW6432Node\Microsoft\Windows\CurrentVersion\Uninstall\* \| Where-Object DisplayName -like 'ScreenConnect Client*' \| Select-Object DisplayName,DisplayVersion`. osquery or an EDR query: `SELECT name, version FROM programs WHERE name LIKE 'ScreenConnect Client%';`. Anything below 26.6.5 is affected.
The flaw is in the client, and updating the console does not update clients. CISA marks forensic triage as required for the KEV entry, so each affected endpoint also needs the evidence step below. | Instance, deployment type, server version, and the count of clients below 26.6.5. | | 0 to 4 hours | Platform engineering | As the interim step, open Administration > Security > Roles, edit each role, and deselect TransferFiles (TransferFilesInSession on legacy versions) for every session group, then save. Review each session group with permissions shown in bold and check Scoped Permissions.
ConnectWise names this as its mitigation; it reduces exposure and does not replace the update. | Roles edited, before and after permission state, and the time. | | 0 to 4 hours | Security lead | Open Administration > Overview and read Latest Eligible Version. On-premises partners must run 25.4 or later to upgrade to 26.6.5; if the license is out of maintenance, renew it first. For cloud, the vendor FAQ says instances are upgraded automatically and the bulletin says no action is required there, but host clients still need to be reinstalled and access agents updated.
The upgrade path decides if today's task is an upgrade or a license step. | Current version, eligible version, and the license date. | | 0 to 4 hours | Security lead | Search the ScreenConnect audit log (Administration > Audit, using the Session Event Type filter) and EDR for file activity no technician made. ConnectWise's session events reference lists SentFiles (sending a file via the host client), CopiedFiles and DraggedFiles (transferring a file from a session by copy and paste or by drag-and-drop), RanFiles (running a file on a remote session) and RanCommand (output from a command run from the Host page). ConnectWise documents no separate event for a transfer made without Host confirmation, so such a transfer shows as one of these events: look for sessions no technician opened, unknown users and unknown source addresses. The audit page notes that the database maintenance plan decides how long data is kept, so export the log first. Huntress reported audit entries with RunFiles or RanFiles showing scripts 1.vbs to 4.vbs run by Process: Guest, and ScreenConnect.WindowsClient.exe launching wscript.exe; these are Huntress's general rogue-client indicators, not tied to CVE-2026-84869 or to ConnectWise's SCKit report. Query: parent image ScreenConnect.WindowsClient.exe with child wscript.exe, cscript.exe or powershell.exe.
Transfer and execution are the two actions the bulletin describes, and these events are where the audit log records them, so each one without a technician behind it needs an owner. | Query text, period, hosts hit, and the audit entries exported. | | 0 to 4 hours | Platform engineering | On every on-premises instance, open the installed extensions list and look for the five extensions ConnectWise named on 13 Nov 2026: Labtech Plus, RMM Plus, RMM+, RMM Plus Password Link and RMM+ Password Link. Disable them and do not reinstall them. Also disable ITBoost Integration, RepairShopr Integration and Guest Session Starter, which ConnectWise disabled in its cloud and marketplace. ConnectWise cannot disable extensions on an on-premises server for you. A listed extension alone does not show compromise, so then review Run Command history, extension API activity, administrative accounts, permission changes and Toolbox records, and investigate endpoints with EDR rather than through ScreenConnect.
ConnectWise reports unauthorized access through these extensions, separate from CVE-2026-84869, and a client update does not remove them. | Per instance: which of the eight extensions were installed, the time each was disabled, and the activity reviewed. | | 4 to 24 hours | Platform engineering | Upgrade on-premises instances to 26.6.5 or later. Afterward reinstall host clients and update access agents, because upgrading the server alone leaves old clients. Verify the version on the endpoint, not only in the console. At scale, build the Windows installer from the upgraded instance and push it with your endpoint tool (Intune, Configuration Manager, Group Policy or an RMM job) as a silent install, for example msiexec /i ScreenConnect.ClientSetup.msi /qn, to every host client, then rerun the step 1 query and list hosts still below 26.6.5 for reinstall by hand. Technician workstations are updated the same way. ConnectWise publishes no deployment command, so this method is ours; the vendor's instruction is only to reinstall host clients and update access agents.
The vendor's remediation covers clients as well as the server, and only a client at 26.6.5 or later is fixed. | Per instance: version before and after, time, and the clients updated. | | 4 to 24 hours | Security lead | Look for persistence. Huntress indicators for rogue ScreenConnect clients: HKCU\Software\Microsoft\Windows\CurrentVersion\Run value WindowsServiceHost pointing to WindowsServiceHost.vbs in AppData, and the staging files %TEMP%\value.txt, map.txt, out.enc and runner.ps1. Look for UltraViewer and Quick Assist installed on hosts where no one asked for them. For the SCKit spread that client 26.6.5 blocks, which ConnectWise describes as moving from an infected endpoint to connecting technician workstations, run the same process query (ScreenConnect.WindowsClient.exe as parent of wscript.exe, cscript.exe or powershell.exe) on the workstations of technicians who connected to hosts running a client below 26.6.5.
Huntress saw these alongside rogue ScreenConnect clients, and removing the client does not remove a Run key. | Hosts checked, indicators found, and the containment action. | | 4 to 24 hours | Security lead | If any host shows signs of compromise, isolate the server, back it up for analysis and keep it offline until it is investigated, rebuilt and patched. Reset passwords, review the audit log and force all technicians to sign back in.
This is the ConnectWise checklist for a suspected compromise. | Server isolated, backup name, rebuild date. | | 24 to 72 hours | Security lead | Review users and roles, remove any user you do not recognize, reset passwords and enable MFA, as the vendor advises after patching. Review authentication, extension installation, administrative, Run Command and Toolbox records, and revoke active web sessions.
Someone who used the tool before the patch may still hold a valid account. | User list with owner for each, MFA status, and the audit export. | | 24 to 72 hours | Executive | Decide whether standing access by outside support staff stays, and whether the tool may be reachable from the internet at all. Record the decision, the owner and the review date.
The patch removes one flaw, and the access model stays the same until someone decides. | Decision, owner, date. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet If no fixed release exists, the mitigations for a remote support tool are to remove the dangerous permission, shrink who can use the tool, and take it offline. The ScreenConnect example shows the first (step 2): technicians lose file transfer in sessions, a feature they use daily. It reduces exposure and does not replace the update. Taking the tool offline stops all remote support, so decide by exposure. If the server is reachable from the internet, the vendor or CISA reports exploitation, and the permission change cannot be made, take it offline; if the server sits behind a VPN and the permission is off, keep it up with the audit log under watch. On an infected instance, isolate and keep the server offline until it is rebuilt and patched. ## CVE-2026-84869 worked through, day by day ConnectWise released ScreenConnect 26.6.5 on 8 Sep 2026 for CVE-2026-84869, a client file-transfer flaw that may allow files to be transferred and executed through an active session without authorization or Host confirmation. The bulletin says "ScreenConnect servers are not impacted." The CNA rates it CVSS 3.1 9.9; ConnectWise's own bulletin rating is Priority 1, High, the label it gives vulnerabilities that are being targeted or at higher risk of being targeted. No public source gives a first attack date or victim count for this CVE itself: the dated reports of exploitation are CISA's SSVC entry of 10 Sep and the KEV listing of 11 Sep. The bulletin, KEV and Huntress give dates, not hours, so the table runs in days. | When | What happened | Source | | --- | --- | --- | | 20 Aug 2026 | Huntress reports its first rogue ScreenConnect client incident, general client activity. | [source](https://www.huntress.com/blog/rogue-screenconnect-installations) | | 3 Sep 2026 | ConnectWise publishes the interim Guest File Transfer advisory with the role mitigation. | [source](https://www.connectwise.com/company/trust/advisories) | | 8 Sep 2026 | ConnectWise releases 26.6.5 and the bulletin for CVE-2026-84869, rated Priority 1 High; NVD publishes the CVE. | [source](https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin) | | 10 Sep 2026 | CISA's SSVC data in the NVD record marks exploitation as active, the earliest dated report of exploitation of this CVE. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-84869) | | 11 Sep 2026 | CISA adds the CVE to KEV, due date 14 Sep 2026, forensic triage required. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-84869) | | 12 Nov 2026, 18:45 UTC | ConnectWise reports unauthorized access to some partner on-premises ScreenConnect environments. It has found no evidence that initial access involved a ScreenConnect vulnerability; the activity appears to abuse independent third-party extensions from the RMM+ family (about 400 partner environments had at least one installed), which expose an API that runs remote commands. Researchers say attackers then deployed a toolkit called SCKit meant to spread from an infected endpoint to connecting technician workstations. The fix for CVE-2026-84869 in client 26.6.5 blocks that spread technique. | [source](https://www.connectwise.com/company/trust/advisories) | | 13 Nov 2026, 21:15 UTC | ConnectWise tells on-premises partners to disable Labtech Plus, RMM Plus, RMM+, RMM Plus Password Link and RMM+ Password Link now, and recommends that all partners also disable ITBoost Integration, RepairShopr Integration and Guest Session Starter, which it disabled in cloud environments and its marketplace. All partners must verify every technician client runs 26.6.5 or later. | [source](https://www.connectwise.com/company/trust/advisories) | | 16 Nov 2026, 20:00 UTC | ConnectWise reports its investigation complete and repeats that the five extensions stay disabled and that every technician client must run 26.6.5 or later. It says researchers continue to report activity against on-premises environments where the extensions remain enabled and ScreenConnect is not updated. | [source](https://www.connectwise.com/company/trust/advisories) | The interim role change was published 5 days before the fix, so an owner who followed the vendor feed could have applied the role change on 3 Sep, 8 days before CISA listed the CVE on 11 Sep. The upgrade on 8 Sep, then client reinstall, would have met the 14 Sep due date. The November notices show the cost of skipping clients: more than 2 months after the release, ConnectWise still asked every partner to verify each technician client, because the SCKit spread step is blocked only by client 26.6.5 and later. The extensions are a separate risk: no client update removes an enabled RMM+ extension, so the extension check belongs in the same first day. ## Related pages - [The artifact repository and CI system playbook for the first 72 hours](/playbooks/artifact-repository/) - [Email gateway playbook: the first 72 hours after a critical flaw](/playbooks/email-gateway/) - [CMS and storefront playbook: the first 72 hours after a critical flaw](/playbooks/cms-and-storefront/) ## Sources - ConnectWise ScreenConnect bulletin, 8 Sep 2026: https://www.connectwise.com/company/trust/security-bulletins/2026-09-08-screenconnect-bulletin (read 19 Nov 2026) - ConnectWise advisories page: https://www.connectwise.com/company/trust/advisories (read 19 Nov 2026) - CISA KEV entry for CVE-2026-84869: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-84869 (read 19 Nov 2026) - NVD record for CVE-2026-84869: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-84869 (read 19 Nov 2026) - Huntress report on rogue ScreenConnect installations: https://www.huntress.com/blog/rogue-screenconnect-installations (read 19 Nov 2026) - ConnectWise ScreenConnect documentation, session events: https://docs.connectwise.com/ScreenConnect_Documentation/Developers/Session_events (read 19 Nov 2026) - ConnectWise ScreenConnect documentation, audit page: https://docs.connectwise.com/ScreenConnect_Documentation/Get_started/Administration_page/Audit_page (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/remote-support-tool/ --- # The artifact repository and CI system playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** An artifact repository holds every build you ship and the tokens that publish them, and the CI system beside it holds the credentials to reach it. JFrog Artifactory CVE-2026-82329 went from advisory on 28 Aug 2026 to about 406,000 attempts in one day. Patch, rotate tokens, then check artifacts. Artifactory and GitLab sit at the point where source becomes software. A flaw in either is rarely about the server itself. The attacker wants what the server can do: mint administrator tokens, read the platform join key and the Access signing keys, add users with SSH keys, drop Groovy user plugins that run commands, or replace an artifact that thousands of builds pull. The Artifactory flaws of 2026 were exploited in exactly that order: an unauthenticated request, an administrator token, a new admin account, then persistence. Wiz saw some actors go from first request to a new admin account in under five minutes. Patching is the hard part, because the patch closes the door and leaves everything the attacker took. An upgrade does not revoke a token already issued, remove an admin user, delete a Groovy plugin or invalidate a stolen signing key. Release trains also differ: Artifactory has six supported trains in the 2026 advisories, and the fixed build for one train is not a fixed build for the next. The CI system adds a second problem. GitLab runners and CI variables hold the tokens that publish to the repository, so a CI compromise becomes a repository compromise even when the repository is patched. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Platform engineering | Copy the logs before touching anything. Collect request*.log from the Artifactory log directory, the reverse proxy or load balancer access log, and on GitLab the Workhorse, api_json, graphql_json and application_json logs. Compute a SHA-256 for each file and store the copies on a host that is not the server.
Attackers on Artifactory truncated request*.log, and an upgrade restart rotates logs, so the evidence is at its fullest now and shrinks with every restart. | A list of file names, sizes, SHA-256 values, the time of copy and who copied them. Note any request log that is empty or missing. | | 0 to 1 hour | Platform engineering | Get the exact version of every self-hosted Artifactory node with GET /artifactory/api/system/version (the version field) and match it to the build list in the upgrade step below. For GitLab, run head -1 /opt/gitlab/version-manifest.txt on Omnibus, or docker exec head -1 /opt/gitlab/version-manifest.txt. Include test, staging and edge instances. Artifactory Cloud and GitLab.com need no action: the vendors patched them.
Every later decision depends on which of six release trains each node runs, and test instances are often the ones exposed to the internet. | One row per node: host, product, version, train, internet-reachable yes or no, owner. | | 0 to 4 hours | Platform engineering | On each Artifactory node, open system.yaml and look under shared, security for additionalJoinKeys (containers and Helm: the variable JF_SHARED_SECURITY_ADDITIONALJOINKEYS). If it is empty you are in the default configuration CVE-2026-82329 needs. Apply JFrog's workaround while the upgrade is prepared: generate a key with openssl rand -hex 16, add it as additionalJoinKeys, keep the existing joinKey entry, restart the Access service or the whole platform deployment. Then block /access/ at the proxy from every network that is not a cluster peer or a trusted admin range.
The workaround closes the empty-key door with one restart, before the upgrade window is agreed, and your existing join key keeps working so cluster traffic is not affected. | The restart time per node, the proxy rule text, and where the new key is stored (it must be as secret as the join key). | | 4 to 24 hours | Platform engineering | Pick one target build that closes all three Artifactory flaws at once. Wiz's list of builds that close CVE-2026-42018 and CVE-2026-82329 is 7.111.21, 7.117.28, 7.125.20, 7.133.29, 7.146.38 or 7.161.20 or later, and JFrog's advisory names 7.133.11 as the only patched build for CVE-2026-42016 (affected: everything before it), so trains 7.111, 7.117 and 7.125 should go to 7.133.29 or later to be sure of closing all three. The per-train targets for the other two bugs are: 7.111 to 7.111.21, 7.117 to 7.117.28, 7.125 to 7.125.20, 7.133 to 7.133.29, 7.146 to 7.146.38 (not 7.146.37: NVD counts everything below 7.146.38 as affected), 7.161 to 7.161.20. Back up first, upgrade, then re-run the version call.
Fixing only the bug in the news leaves the other two. The chain of CVE-2026-42018 and CVE-2026-42016 turned an unauthenticated request into an admin token in two calls, and the fixes for those shipped on different builds. | Version before and after for each node, the backup location, and the time the service came back. | | 4 to 24 hours | Security lead | Search the proxy and application logs back to 28 Aug 2026. High-confidence evidence of CVE-2026-82329 is a POST to /access/api/v1/registry/join (also /join/router and /join/router?override=true) that returned HTTP 201 with a token whose key ID is e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855, the SHA-256 of an empty string. If your logs do not record response bodies, list every 200 or 201 on those paths from an address that is not a cluster peer. For the CVE-2026-42018 chain, look for a 401 on the bare AWS token path followed by a 200 on a variant of it from the same client, then POST /access/api/v1/tokens, then requests whose actor is token:anonymous creating users.
The result decides the next step: no hit means rotate as a precaution, a hit means you run an incident and the rest of this table is evidence work. | The query text, the log range searched, each hit with source address, time and status, and the count of hits (zero is a result worth writing down). | | 4 to 24 hours | Platform engineering | For GitLab, upgrade the self-managed instance to 19.2.8, 19.3.5 or 19.4.2 or later for CVE-2026-97229; 19.4.1 is not enough because its earlier fix was incomplete. Until then limit who can create accounts and personal access tokens (Admin, Settings, General, Sign-up restrictions). Check the Workhorse log for a 502 with 'badgateway: failed to receive response: EOF' on POST /api/v4/projects/:id/ci/lint or /api/graphql, with a large read_bytes value and a correlation_id missing from api_json.log and graphql_json.log. GitLab calls that the highest-confidence signal.
The CI regex flaw gives code execution as the git user on the server that stores your pipeline secrets, and CISA lists it as exploited with a public Metasploit module merged on 30 Oct 2026. | GitLab version before and after, the sign-up setting before and after, and every Workhorse hit with its correlation_id. | | 4 to 24 hours | Security lead | Rotate what an unauthenticated administrator could have read. Replace the platform join key. Revoke every access token issued since 28 Aug 2026 on any node that was reachable while below its fixed build, then reissue credentials to the CI jobs that need them. If you find signs of entry, or Access private.key may have been read, reset the Access token certificate: it invalidates every token and needs a planned window, but a patch does not revoke forged tokens.
Fastly and Wiz both say patching does not revoke tokens already issued, and the attackers took join keys and created long-lived tokens in the first minutes. | The token list before revocation, the time of revocation, and the list of services that had to be reissued. | | 24 to 72 hours | Security lead | Hunt for persistence the upgrade left in place. Users: list all users and compare with the identity provider; names seen in 2026 were Nxploited_ plus 3 characters, labadmin_ plus hex, 0xTerror, svc_ plus 8 characters, and look-alikes such as jfrog-distribution, jfrog-insight and repo-service; also check SSH keys attached to users. Files: ls -la $JFROG_HOME/var/etc/artifactory/plugins/ for rce.groovy, cmd.groovy, jwtToken.groovy, uploadFile.groovy, JFrogImage.groovy, metrics.groovy, httpSession.groovy; a payload at /tmp/.z with SHA1 513a907b69edffc3cb77a494da395178d21ef9bd; binaries in /dev/shm, /tmp and /var/tmp. If the plugin API is broken, or registry/join, api/plugins or systemUpload requests are rejected with HTTP 400, an in-memory filter named traceFilter may be installed: restart the service from a clean state.
These artifacts survive an upgrade and are how an attacker comes back after you think the incident is closed. | Each unexpected user, key, plugin and file with its timestamp and hash, and the decision taken on it. | | 24 to 72 hours | Application owner | Find what changed in your repositories during the exposure window with an AQL query posted to /artifactory/api/search/aql, for example items.find({"repo":"","modified":{"$gt":"2026-08-28T00:00:00.000Z"}}).include("name","path","sha256","modified_by"), and flag every artifact whose modified_by is not your CI publishing identity. Rebuild the flagged artifacts and the three most widely pulled ones from source on a clean runner and compare SHA-256 values. Hold the next release until they match.
A repository admin token can overwrite a released file, and the artifact looks the same until its checksum is compared with a clean rebuild. | The AQL output, the rebuilt hashes beside the stored hashes, and the release hold and its release time. | | 24 to 72 hours | Security lead | Rotate the secrets CI holds for the repository: the CI/CD variables that publish to Artifactory, runner registration and authentication tokens, deploy keys and access tokens, and gitlab-secrets.json if there are signs of code execution on GitLab. Replace long-lived Artifactory tokens in CI with short-lived ones scoped to one repository. Hunt files owned by git in /tmp and unexpected child processes of Puma or Sidekiq.
CI is the way a GitLab compromise reaches the repository, so the repository's tokens are only clean once CI's copies are replaced. | Each secret rotated with the time, the job that uses it, and the owner who confirmed the job still publishes. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet JFrog gave no workaround for CVE-2026-42016 or CVE-2026-42018, and GitLab gave none for CVE-2026-97229. The only controls then are exposure controls. Restrict the web and Access paths to trusted networks or a VPN; the cost is that remote developers, external consumers and CI outside the allowed path lose access. On GitLab, limit sign-ups and token creation, which narrows who can reach the flaw without removing it. For CVE-2026-82329 JFrog did publish a workaround (the additionalJoinKeys step), and Fastly Next-Gen WAF customers could enable a virtual patch, which is a compensating control and not a fix. Take the service offline when three things are true together: the node is below its fixed build, it cannot be moved behind an allow-list within hours, and the logs show hits or are missing. CISA's entry for CVE-2026-82329 tells you to follow vendor mitigations and to discontinue use of the product if mitigations are unavailable. Offline costs failed builds and failed deploys, so before you cut it, check that CI can pull dependencies from a remote-repository cache or a read-only mirror. Give the decision to a named person, with the time recorded. ## CVE-2026-82329 worked through, hour by hour JFrog published the advisory for the Artifactory join-key bypass on 28 Aug 2026. On a default install the Access service accepted an empty string as the cluster join key, so an unauthenticated caller could obtain an administrator token. CVSS base score 9.8. The table follows one self-hosted instance from advisory to mass exploitation, using Fastly's daily counts and the CISA entry. Fastly and JFrog give dates and daily counts, not hours, so the table keeps days; Wiz's change log gives the only hour-level times. | When | What happened | Source | | --- | --- | --- | | 28 Aug 2026, Day 0 | JFrog publishes the advisory with fixed builds 7.111.21, 7.117.28, 7.125.20, 7.133.29, 7.146.38 and 7.161.20 and the additionalJoinKeys workaround. NVD publishes the record. Fastly sees no exploitation that day. | [source](https://docs.jfrog.com/releases/docs/jfrog-security-advisories) | | 29 and 30 Aug 2026, Day 1 and 2 | Fastly records only single-digit probes on each day. | [source](https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity) | | 31 Aug 2026, Day 3 | About 75,000 attempts, of which roughly 73,000 came from offensive-security vendors and research services (nearly 98%). About 72 hours after the advisory. | [source](https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity) | | 1 Sep 2026, Day 4 | A public exploit appears. Fastly counts just above 171,000 attempts. Wiz's observed successful exploitation runs from this day to 8 Sep 2026. | [source](https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity) | | 2 Sep 2026, Day 5 | CISA adds the CVE to KEV with a due date of 5 Sep 2026 and forensic triage marked Yes. Fastly counts its peak, about 406,000 attempts. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-82329) | | 3 Sep 2026, Day 6 | Fastly publishes its analysis and a WAF virtual patch rule for its Next-Gen WAF customers. | [source](https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity) | | 10 Sep 2026, Day 13 | Wiz publishes its report of successful exploitation: configuration theft, new admin accounts, long-lived tokens, join key theft, attached SSH keys and Groovy plugins. It corrects the HTTP method for this CVE and the fixed versions for CVE-2026-42018 at 15:00 UTC on 11 Sep 2026, and adds the Groovy plugin section at 19:00 UTC on 17 Sep 2026. | [source](https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201) | The playbook's first four steps fit inside the 72 hours before the first large wave. Logs copied and the additionalJoinKeys workaround applied in the first 4 hours, on 28 Aug, closes the empty-key door about three days before the 75,000-attempt day and about four days before the public exploit. An instance upgraded to its train build by 31 Aug was patched before the first success in Wiz's observed window, which opens 1 Sep. An instance that waited for the KEV listing acted on 2 Sep, day 5, the day of the 406,000 peak (5.4 times the 31 Aug volume), and for it the 201 search, the token rotation and the plugin hunt in the later steps decided whether the instance was safe. ## Related pages - [CMS and storefront playbook: the first 72 hours after a critical flaw](/playbooks/cms-and-storefront/) - [The remote support tool playbook for the first 72 hours](/playbooks/remote-support-tool/) - [The Windows operating system playbook for the first 72 hours](/playbooks/windows-endpoint-os/) ## Sources - JFrog security advisories: https://docs.jfrog.com/releases/docs/jfrog-security-advisories (read 19 Nov 2026) - Fastly on CVE-2026-82329: https://www.fastly.com/blog/cve-2026-82329-jfrog-artifactory-authentication-bypass-exploitation-activity (read 19 Nov 2026) - Wiz Research on Artifactory CVE-2026-42018 and CVE-2026-42016: https://www.wiz.io/blog/artifactory-under-attack-in-the-wild-exploitation-of-cve-2026-42016-cve-2026-4201 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-82329: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-82329 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-97229: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-97229 (read 19 Nov 2026) - NVD record for CVE-2026-82329: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-82329 (read 19 Nov 2026) - NVD record for CVE-2026-42016: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-42016 (read 19 Nov 2026) - NVD record for CVE-2026-42018: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-42018 (read 19 Nov 2026) - GitLab patch release 19.4.2: https://docs.gitlab.com/releases/patches/patch-release-gitlab-19-4-2-released/ (read 19 Nov 2026) - JFrog documentation: Get Artifactory Version API: https://docs.jfrog.com/administration/reference/getartifactoryversion (read 19 Nov 2026) - NVD record for CVE-2026-97229: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-97229 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/artifact-repository/ --- # CMS and storefront playbook: the first 72 hours after a critical flaw Playbook, sources read 19 Nov 2026 **The short answer.** Content and commerce platforms run public PHP code with writable folders, so one flaw gives a web shell and often payment data. This playbook covers WordPress CVE-2026-119977 (fixed in 7.1.4 and backports to 4.7.39) and Adobe Commerce CVE-2026-75650 (hotfix VULN-39341, not a release), hour by hour. A CMS or storefront is a large body of PHP code that anonymous visitors reach by design, on a host where the web user can write files. Flaws are exploited by getting attacker text onto disk and getting the application to run it: a log or error report that stores a request, a predictable upgrade folder, a template engine that builds objects from visitor input. The attacker gains code execution as the web user, then the database credentials, the admin session store, the payment integration keys and, on a shop, the checkout page itself. Two things make patching harder than it looks. The fix may not be a version number: Adobe's fix for CVE-2026-75650 is a composer hotfix applied to every 2.4 release, so a version read says nothing about whether you are safe. And the upgrade can itself be the exposure: WordPress CVE-2026-119977 can be exploited only while a core upgrade is running, and the upgrade that installs the fix still runs the old, vulnerable updater. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Step 1. Inventory every WordPress install, including staging copies, agency-run sites and multisite networks, and read each version: wp core version, or grep wp_version wp-includes/version.php. WordPress CVE-2026-119977 affects 4.7.0 through 7.1.3. First fixed release per branch: 7.1.4, 7.0.8, 6.9.11, 6.8.12, 6.7.11, 6.6.11, 6.5.14, 6.4.14, 6.3.14, 6.2.15, 6.1.16, 6.0.18, 5.9.20, 5.8.19, 5.7.21, 5.6.23, 5.5.24, 5.4.25, 5.3.27, 5.2.30, 5.1.28, 5.0.31, 4.9.35, 4.8.34 and 4.7.39 (all released 28 Oct 2026; 7.1.5, 7.0.9 and 6.9.12 followed on 18 Nov 2026 and are the newest on their branches). Any site older than the backport for its branch, or on a branch below 4.7, is outside the vendor's fix.
The fixed version differs by branch, so a single 'update to latest' instruction fits only sites you control. | Site, branch, exact version, who updates it, whether automatic core updates are on. | | 0 to 1 hour | Security lead | Step 2. Inventory Adobe Commerce, Adobe Commerce B2B and Magento Open Source stores. Read bin/magento --version, and for the hotfix run vendor/bin/magento-patches -n status \| grep "39341\\|Status" from the Magento root; the line for 39341 must show Applied. Affected: Adobe Commerce, Magento Open Source and Commerce B2B 1.3.3 to 1.5.3, on 2.4.4 to 2.4.9 at every patch level through the 2026-aug releases (Sansec: "Every version from 2.4.4 up to and including 2.4.9 is affected"; Adobe tested the hotfix on exactly that range). Older versions are affected in practice too: the NVD record built from Adobe's data lists 2.4.6 to 2.4.9 and every earlier version, and the unsupported branches 2.2, 2.3 and 2.4.0 to 2.4.3 get no Adobe patch. The fix is Adobe hotfix VULN-39341 from APSB26-146, a composer patch, not a new release number.
The version cannot tell you the answer, because a store on 2.4.6-p15 with the July and August patches was the first victim Sansec saw. | Store, product line, patch-status line for 39341 (Applied or not), public yes or no. | | 1 to 2 hours | Platform engineering | Step 3. Block web access to the WordPress staging folder before any upgrade runs, including automatic ones. nginx: location ^~ /wp-content/upgrade/ { deny all; } . Apache: a Require all denied rule for the wp-content/upgrade directory. Reload the web server, then from outside run curl -sI https://your-site.example/wp-content/upgrade/ and expect 403 or 404. This rule comes from comparing the 7.1.3 and 7.1.4 code, not from a vendor statement; 7.1.4 itself adds an index.php to the folder and a 16 character random suffix to the working folder.
The upgrade that installs the fix still creates the predictable folder, so the block has to exist first. | Server config file and line, the curl status code before and after. | | 1 to 2 hours | Platform engineering | Step 4. Put a WAF virtual patch on every Commerce and Magento store: Fastly offers one for its Next-Gen WAF and Sansec Shield blocks the variants it has seen. Treat it as compensating only. Take a disk snapshot, and copy var/log, var/report, a listing of pub/media and the cron spool file before the hotfix.
Exploitation began 4 Sep 2026, three days before the hotfix existed, and the vendors say payloads changed several times a day, so the WAF buys time and the evidence copy keeps the proof. | WAF rule id and time active; snapshot id; where the copies are stored. | | 2 to 4 hours | Security lead | Step 5. Hunt a Commerce or Magento host for the indicators Sansec lists. Processes and cron: crontab -l \| grep -i gvfsd; ps -eo pid,comm,args \| grep -iE 'kworker\|fc-cache\|chronyd' (a chronyd or fc-cache process whose parent is PID 1 is suspect; also read the cron spool file, since an empty crontab is not proof). Files: grep -r 'crontab command not allowed' /var/log/ ; grep -ril 'x_trace_' var/report/ ; find -L pub/media -type f -name '*.php' (a sync_<10 hex>.php under pub/media/catalog/product/cache/ss_<10 hex>/ is a web shell, and a modified vendor/magento/framework/App/View.php that loads remote code is another sign). Network: C2 99[.]84[.]67[.]186 on 443, NTP-shaped beacons to ntp[.]timesync[.]to and ntp[.]timesysnc[.]net, download host 209[.]141[.]43[.]95, second C2 213[.]111[.]146[.]27.
Attackers were in stores before the hotfix, so patching alone leaves an existing web shell running. | Each hit with path, owner, modification time; any process with its parent PID. | | 4 to 24 hours | Platform engineering | Step 6. Apply Adobe hotfix VULN-39341: download VULN-39341-composer-patches.zip from repo.magento.com, apply it as a composer patch on staging, deploy (static content and cache steps mean downtime), then re-run the vendor/bin/magento-patches status line and expect Applied. Stores on 2.2, 2.3 or 2.4.0 to 2.4.3 get no Adobe patch: upgrade to a supported release, or test Scandiweb's third-party backports (41 older releases, 2.2.0 to 2.4.3-p3) on a staging copy first, because Sansec states it has made no review of them.
Everything before this step only reduced exposure; the hotfix removes the flaw. | Time deployed per store, the status line, the output of the first working checkout test. | | 4 to 24 hours | Platform engineering | Step 7. Update each WordPress site to the fixed release for its branch, for example wp core update --minor, and let automatic updates run once the staging block from step 3 is in place. After the upgrade, run wp core verify-checksums, wp user list --role=administrator and compare with the list you hold, ls -la wp-content/upgrade/ for leftover folders named after the package (wordpress- with no random suffix), and a wp-config.php or any .php file in that folder. List new files under wp-content/uploads and wp-content/plugins. Search the web server log for requests to /wp-content/upgrade/ in the minutes around each past upgrade.
WordPress.org names no indicators, so file and account differences from a known-good state are the only evidence there is. | Version before and after, checksum result, administrator list, any leftover folder. | | 24 to 72 hours | Platform engineering | Step 8. On every Commerce and Magento store, rotate what the attacker could read, as Adobe recommends: the encryption key and every credential it protected (admin passwords, REST, SOAP and GraphQL integration tokens, OAuth client secrets, payment gateway API credentials, database credentials, SSH and deploy keys, third-party extension API keys). Rotate at the source, not only in Magento, because changing the encryption key does not invalidate what was already read. Do it after the hotfix and the host hunt, so the new secrets are not handed to a shell that is still there.
Secrets read before the patch stay valid after it. | Each credential, where it was rotated, time, and who confirmed it works. | | 24 to 72 hours | Application owner | Step 9. Test checkout, login, search, plugin and payment email flows on the fixed versions, and turn automatic core updates back on for WordPress if you turned them off. Check the mail log for bursts of the 'Payment Transaction Failed Reminder' email, which the exploit triggers, remembering that real declines send the same message.
Fixes on a shop break checkout in ways staging does not show, and the email burst is the only traffic-level sign of the attack. | Test cases passed, the declined-payment email count per day for the exposure window. | | 24 to 72 hours | Communications | Step 10. If a web shell or modified file was found on a store that serves checkout, tell your payment processor or acquirer and your counsel the same day and agree on the customer notice. Name the window from the earliest file time, not from the patch date.
Processors usually require prompt notice of a suspected compromise on a card-handling host, and the earliest file time sets how many customers are in scope. | Date the host was first suspect, systems in scope, the notice sent. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet For Adobe Commerce, the period before the hotfix was a no-fix case: exploitation from 4 Sep, hotfix on 7 Sep. The only mitigations were a WAF virtual patch, host monitoring and evidence collection. The WAF costs false positives on admin template edits. Moving sessions to Redis or the database does not stop the attack, and switching off outbound mail does not either, since the attack works when mail delivery fails. Unsupported branches still have no Adobe patch, so a third-party backport or an upgrade is the only route. For WordPress the mitigation is the staging block plus restricting who and what can start an upgrade. It costs nothing for visitors. Take a store offline when indicators are found on a checkout host, or when the hotfix cannot be applied and the store runs an unsupported branch, because CISA's KEV action says to stop using the product if mitigations are unavailable. A maintenance page for a day costs less than a shell on the payment page. ## CVE-2026-75650 and CVE-2026-119977 worked through, hour by hour and then day by day Adobe Commerce and Magento CVE-2026-75650 is a server-side template injection rated CVSS 10.0, exploited before Adobe's hotfix existed; its rows are timed in UTC hours. WordPress CVE-2026-119977 is an upgrade-window flaw rated CVSS 8.1 with a fix on the day of disclosure and a KEV listing six days later; its rows are given by day. | When | What happened | Source | | --- | --- | --- | | 4 Sep 2026, 22:20 UTC | Sansec records the first confirmed exploitation of what it names StyleSmuggler. At 23:10 its scanner flags the implant on unrelated stores. The first victim ran 2.4.6-p15 with the July and August 2026 patches. | [source](https://sansec.io/research/stylesmuggler-0day) | | 5 Sep 2026, 07:15 UTC | Sansec Shield starts blocking the attacks, and Sansec publishes the analysis the same day. | [source](https://sansec.io/research/stylesmuggler-0day) | | 7 Sep 2026, 20:20 UTC | Adobe publishes APSB26-146 at priority 1 with hotfix VULN-39341 (time per Sansec). It ships as a composer patch, not a release. The bulletin's data appears in the NVD record below. | [source](https://sansec.io/research/stylesmuggler-0day) | | 7 Sep 2026, 21:17 UTC | NVD publishes the record from Adobe's CNA data: CVSS 3.1 10.0, CWE-1336. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-75650) | | 8 Sep 2026 | CISA adds the CVE to KEV, forensic triage required, due date 11 Sep 2026. SSVC: exploitation active, automatable yes, technical impact total. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-75650) | | 14 Sep 2026, 17:00 UTC | Sansec updates the article with further indicators. | [source](https://sansec.io/research/stylesmuggler-0day) | | 16 Sep 2026 | Fastly publishes its analysis, a Nuclei template and a virtual patch, and says operators changed payloads several times a day. | [source](https://www.fastly.com/blog/what-is-style-smuggler-cve-2026-75650) | | 28 Oct 2026 | WordPress 7.1.4 is released with fixes for the upgrade-process flaw (reported by Umut Bayram). Backports to every branch back to 4.7 (7.0.8 down to 4.7.39) ship the same day. | [source](https://wordpress.org/news/2026/10/wordpress-7-1-4) | | 31 Oct 2026, 06:16 UTC | NVD publishes CVE-2026-119977 from the Wordfence CNA record: CVSS 3.1 8.1, attack complexity High, CWE-367. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-119977) | | 3 Nov 2026 | CISA adds CVE-2026-119977 to KEV with a due date of 6 Nov 2026 and forensic triage Yes. This listing is the earliest dated public report of exploitation; CISA names no first exploitation date, actor or indicators, and WordPress.org lists none. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-119977) | | 18 Nov 2026 | WordPress 7.1.5, 7.0.9 and 6.9.12 are released as the newest builds on their branches. | [source](https://wordpress.org/download/releases/) | From the first confirmed exploitation at 22:20 on 4 Sep to Adobe's hotfix at 20:20 on 7 Sep is 70 hours with no vendor fix. An owner who started on Sansec's article on 5 Sep would have read the patch status (step 2), put a WAF rule and an evidence copy in place (step 4) and run the host hunt (step 5) inside the first four hours, roughly two and a half days before the hotfix. Applied within 24 hours of its publication at 20:20 UTC on 7 Sep, the hotfix was in by 20:20 UTC on 8 Sep, the day CISA added the entry to KEV, and credential rotation fits before the 11 Sep due date, three days after listing. The step-2 patch-status line is the one that would have caught the patched-version trap: the 2.4.6-p15 store looked current and was not safe. The WordPress flaw runs the other way: the fix existed on day 0 (28 Oct) and the KEV listing came on day 6 (3 Nov), with the due date on day 9 (6 Nov). A team that ran the inventory and the staging block (steps 1 and 3) inside the first two hours of 28 Oct and updated within 24 hours had a fixed fleet five days before the listing. A site left on its old version, or one that let an automatic update run without the block, was exposed for those six days. ## Related pages - [The Windows operating system playbook for the first 72 hours](/playbooks/windows-endpoint-os/) - [The artifact repository and CI system playbook for the first 72 hours](/playbooks/artifact-repository/) - [The Linux server and SSH playbook for the first 72 hours](/playbooks/linux-server-ssh/) ## Sources - WordPress 7.1.4 release note: https://wordpress.org/news/2026/10/wordpress-7-1-4 (read 19 Nov 2026) - WordPress.org releases page: https://wordpress.org/download/releases/ (read 19 Nov 2026) - CISA KEV entry for CVE-2026-119977: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-119977 (read 19 Nov 2026) - NVD record for CVE-2026-119977: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-119977 (read 19 Nov 2026) - WordPress core class-wp-upgrader.php, 7.1.4 tag: https://raw.githubusercontent.com/WordPress/wordpress-develop/7.1.4/src/wp-admin/includes/class-wp-upgrader.php (read 19 Nov 2026) - Sansec: StyleSmuggler, CVE-2026-75650: https://sansec.io/research/stylesmuggler-0day (read 19 Nov 2026) - Fastly: Style Smuggler, CVE-2026-75650: https://www.fastly.com/blog/what-is-style-smuggler-cve-2026-75650 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-75650: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-75650 (read 19 Nov 2026) - NVD record for CVE-2026-75650, carrying Adobe's APSB26-146 data: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-75650 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/cms-and-storefront/ --- # The Windows operating system playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Three Windows privilege flaws were exploited when Microsoft shipped their fixes: CVE-2026-68820 on 11 Aug 2026, CVE-2026-85880 and CVE-2026-81963 on 8 Sep 2026. The work is matching every host to a build floor, restarting it and reading the build back. The build numbers, KB articles and deadline settings are below. Windows zero-days in this set arrive with the fix already out. Microsoft marked each flaw Exploited:Yes in the monthly release that carried the update, so the first hour of the clock is Patch Tuesday itself, with a cumulative update waiting and a restart pending. All three are local privilege escalations (CVSS vector AV:L, PR:L): the attacker already runs code on the host and the flaw takes them to SYSTEM. CVE-2026-85880 in ALPC is built to leave an AppContainer, the sandbox that holds browser content and Store apps. CVE-2026-68820 in the WinSock driver afd.sys is a race condition, which Microsoft rates High attack complexity and still marks exploited. CVE-2026-81963 in the Update Stack is a link-following flaw that is listed only for Windows 11 and Windows Server 2025. What makes this hard is bookkeeping, not discovery. The fix is a cumulative update with a different build number for each of a dozen Windows releases. The KB article number proves little, because a later cumulative update carries the same fix under another KB. And the update does nothing until the host restarts. This page covers the operating system. The browser that may carry the attacker's first foothold is covered in the endpoint fleet playbook. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Read Microsoft's own exploit flag, then CISA's entry. Download the month's CVRF file with curl -s https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep -o 2026-Sep.xml (about 36 MB for September 2026) and find the CVE's Exploit Status threat. For CVE-2026-85880, CVE-2026-81963 and CVE-2026-68820 it reads "Publicly Disclosed:No;Exploited:Yes;Latest Software Release:Exploitation Detected". Then read dateAdded, dueDate and forensicTriage for the same CVE in the KEV feed at https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json. All three were added the day Microsoft published, with a due date 14 days later and forensicTriage No.
The exploited flag decides whether the restart happens this week or in the monthly cycle, and Microsoft publishes it together with the fix, so the clock starts at release. | CVE, the Exploit Status string, KEV dateAdded, dueDate and forensicTriage, and the time you read them. | | 0 to 4 hours | Platform engineering | Fix one build floor for each Windows release before anyone queries a host. A cumulative update carries earlier fixes, so the September 2026 update for a release clears every one of the three flaws that applies to it. Floors: Windows 10 22H2 10.0.19045.7725 and 21H2 10.0.19044.7725 (both KB5122878); Windows 10 1809 and Server 2019 10.0.17763.9245 (KB5122876); Windows 10 1607 and Server 2016 10.0.14393.9512 (KB5123099); Server 2022 10.0.20348.5622 (KB5122882); Server 2012 R2 6.3.9600.23398 (KB5123066); Server 2012 6.2.9200.26349 (KB5123065); Windows 11 23H2 10.0.22631.7582 (KB5122880); Windows 11 24H2 10.0.26100.9445 and 25H2 10.0.26200.9445 (both KB5124008); Windows 11 26H1 10.0.28000.2954 (KB5124012); Server 2025 10.0.26100.33438 (KB5122871). Windows 11 24H2 and Server 2025 share build 26100 with different revisions, so compare each to its own row. CVE-2026-85880 is listed for Windows 10 and Server 2012 to 2022, CVE-2026-81963 for Windows 11 and Server 2025, and CVE-2026-68820 for all of them.
Counting patched hosts by KB article undercounts, because a host on a later cumulative update shows a different KB; build and revision compare cleanly. | The floor table with the date you checked it against the MSRC file. | | 0 to 4 hours | Platform engineering | Read the build and the restart state off every host in one pass: Invoke-Command -ComputerName (Get-Content .\hosts.txt) -ScriptBlock { $v = Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion'; [pscustomobject]@{Host=$env:COMPUTERNAME; Type=$v.InstallationType; Release=$v.DisplayVersion; Build="$($v.CurrentBuild).$($v.UBR)"; RebootPending=(Test-Path 'HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\WindowsUpdate\Auto Update\RebootRequired')} } \| Export-Csv builds.csv -NoTypeInformation. Match each row to its floor by Type and CurrentBuild, then compare the revision after the dot. Intune or ConfigMgr inventory gives the same two numbers if you cannot reach hosts with remoting.
The revision moves only when the update is installed and the host has restarted, so build plus RebootPending separates a patched host from one that is waiting. | builds.csv with the run time, and the count of hosts below the floor for each release. | | 0 to 4 hours | Security lead | Order the hosts by how many people can run code on them. First: Remote Desktop session hosts, VDI pools, jump hosts, shared lab and kiosk machines and build agents. Second: servers that open files from outside, such as mail, upload and document-conversion servers. Third: administrator workstations. Last: single-user laptops. List who can sign in with net localgroup "Remote Desktop Users" and qwinsta /server:HOST.
All three flaws need code on the host first, so the hosts where the most people or the most untrusted content reach a local session are the ones where an exploited flaw is easiest to use. | A tier number for each host, added as a column in builds.csv. | | 4 to 24 hours | Platform engineering | Read Microsoft's Known Issue entry for your floor KB before the pilot. The September CVRF lists one for 9 of the 10 September KB articles (all except KB5123065 for Server 2012). Pilot on IT machines plus one host from each tier for 4 hours, then release the first wave. Do not hold a wave for a cosmetic issue, and do hold it for anything that stops sign-in, printing, a line-of-business app or the agent that reports patch state.
A bad cumulative update found at hour 6 costs a pilot; found at hour 30 it costs the fleet. | The Known Issue text for each floor KB, the pilot hosts, and the time the first wave was released. | | 4 to 24 hours | Platform engineering | Force install and restart on workstations. With Intune, create an expedite policy for the floor update; it needs Windows Pro, Pro Education, Enterprise or Education (not LTSC), telemetry set to at least Required, and the wlidsvc service running. Without it, use Windows Update deadlines: set ConfigureDeadlineForQualityUpdates (range 0 to 30 days) to 1 and ConfigureDeadlineGracePeriod (range 0 to 7, default 2) to 0. Group Policy path: Computer Configuration > Administrative Templates > Windows Components > Windows Update > Manage end user experience > Specify deadline for automatic updates and restarts. Microsoft: "After the deadline passes, restarts will occur regardless of active hours and users won't be able to reschedule." These two values are our suggestion for an exploited flaw, not Microsoft's.
Without a deadline, automatic restarts wait for hours outside active hours, and Microsoft's default is seven days before the user is even told a restart is required. | The policy name and values, the groups they apply to, and the time the policy first reached a device. | | 4 to 24 hours | Platform engineering | Handle servers by hand, not by deadline. Microsoft's restart guidance for servers is Configure Automatic Updates option 7, notify to install and notify to reboot. Take the tier-one servers from the ordering step first. Patch one node of a cluster at a time, or let Cluster-Aware Updating do it with Invoke-CauRun -ClusterName NAME -MaxFailedNodes 1 -RequireAllNodesOnline. Restart domain controllers one at a time and run repadmin /replsummary before the next.
A forced restart on a server in business hours is an outage, and the shared servers are the ones that most need the update. | The change ticket for each server, the restart time and the build read back after the restart. | | 24 to 72 hours | Security lead | Rerun the build query every 4 hours and sort the hosts below the floor into two groups. RebootPending true means the update is installed and the host is waiting: contact the owner and schedule the restart. RebootPending false means the update never installed: check that the host reported in, that its update source approves the KB, and that the disk has room. For laptops that stay off the network, create one Intune compliance policy for each release, assigned by an Entra dynamic device group such as device.deviceOSVersion -startsWith "10.0.19045", with Minimum OS version set to that release's floor, and require a compliant device in Conditional Access.
A single minimum such as 10.0.19045.7725 would pass every Windows 11 device at any revision, because 26100 is a larger number than 19045. | The straggler list with the group, the owner and the next action for each host. | | 24 to 72 hours | Security lead | Write an exception for every host that cannot reach its floor: a release missing from the floor table, a host under a change freeze, or an application that breaks on the update. Each exception names the owner, the date the host will reach the floor, and the compensating control from the no-fix section. Note that the KEV due date is 14 days after the fix, so the exceptions list is what you hand over on day 3 and not on day 14.
Microsoft lists no workaround, so an unpatched host with no written control is an unmanaged risk with a date on it. | The exceptions list with owner, date and control; the final percentage at or above floor for each tier. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet For these three flaws Microsoft lists no workaround, and neither Microsoft nor CISA publishes a log signature, file name or network indicator. When a host cannot take the update, the only levers cut the supply of local code. Remove who can sign in: list the Remote Desktop Users group and take out accounts that do not need it, with Remove-LocalGroupMember. Block unapproved programs with Windows Defender Application Control or AppLocker. For CVE-2026-85880, keep browsers and document readers current, because the attack starts inside an AppContainer. The costs are real: people lose access to shared hosts, and unsigned line-of-business software stops starting under application control. Decide by who can run code, not by the severity label. Take a host offline, or close it to all but a named administrator, when several unrelated users or untrusted files reach a local session and the update cannot be installed inside the KEV window of 14 days. Leave a server online when nobody signs in interactively and it opens no outside files, and patch it in the next window inside the due date. Microsoft's Exploited:Yes flag means someone already has the second stage of an attack, so the open question is whether they can get the first stage on that host. ## CVE-2026-68820, CVE-2026-85880 and CVE-2026-81963 worked through, hour by hour Two Patch Tuesdays four weeks apart carried three exploited privilege flaws. August 2026 fixed the WinSock driver flaw CVE-2026-68820 for every supported release. September 2026 fixed the ALPC flaw CVE-2026-85880 for Windows 10 and Server 2012 to 2022, and the Update Stack flaw CVE-2026-81963 for Windows 11 and Server 2025. The times below start at the CVRF release time, which the file gives as 07:00 with no time zone. | When | What happened | Source | | --- | --- | --- | | 11 Aug 2026, 07:00 | Microsoft's August release lists CVE-2026-68820 with Exploited:Yes and Publicly Disclosed:No. It carries 13 fixed KB articles, from KB5120386 for Server 2012 to KB5121000 for Windows 11 26H1. Windows 10 22H2 needs 10.0.19045.7663 (KB5120249). | [source](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug) | | 11 Aug 2026 | CISA adds CVE-2026-68820 to the KEV catalog the same day, with forensicTriage No and a due date of 25 Aug 2026. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | | 25 Aug 2026 | KEV due date for CVE-2026-68820, 14 days after the fix. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | | 8 Sep 2026, 07:00 | Microsoft's September release (999 CVEs) lists CVE-2026-85880 and CVE-2026-81963 as Exploited:Yes, both with revision 1.0. Windows 10 22H2 now needs 10.0.19045.7725 (KB5122878); Windows 11 24H2 needs 10.0.26100.9445 (KB5124008). | [source](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep) | | 8 Sep 2026 | CISA adds both September flaws to KEV the same day with forensicTriage No and a due date of 22 Sep 2026. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | | 11 Sep 2026, 07:00 | Hour 72 of the clock started on 8 Sep. Microsoft publishes revision 1.1 of CVE-2026-81963, which updates an acknowledgement and is marked informational, so nothing needs redoing. | [source](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep) | | 16 Sep 2026, 07:00 | Microsoft revision 1.1 of CVE-2026-85880 corrects the temporal Exploit Code Maturity value in the CVSS vector. The change is informational and the base score stays 7.8. | [source](https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep) | | 22 Sep 2026 | KEV due date for CVE-2026-85880 and CVE-2026-81963, 14 days after the fix. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | The September floor for Windows 10 22H2 is 62 revisions above the August floor (7725 against 7663), and for Windows 11 24H2 it is 277 above (9445 against 9168), so a host that took the August update still needed the second restart. A fleet that waits for each KEV due date spends 14 days, or 336 hours, on an exploited build after each fix: 28 days across the two months. Following the steps above, the restart is forced within about a day on workstations and the straggler list is closed by hour 72, which leaves 6 days across the two months and finishes 264 hours (11 days) before the September due date. ## Related pages - [The Linux server and SSH playbook for the first 72 hours](/playbooks/linux-server-ssh/) - [CMS and storefront playbook: the first 72 hours after a critical flaw](/playbooks/cms-and-storefront/) - [The AI infrastructure playbook for the first 72 hours](/playbooks/ai-infrastructure/) ## Sources - Microsoft Security Response Center CVRF, August 2026 release (CVE-2026-68820: exploit status, fixed builds, KB articles): https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Aug (read 19 Nov 2026) - Microsoft Security Response Center CVRF, September 2026 release (CVE-2026-85880 and CVE-2026-81963: exploit status, fixed builds, KB articles, known issues, revision history): https://api.msrc.microsoft.com/cvrf/v3.0/cvrf/2026-Sep (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities feed, catalog version 2026.11.18: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026) - Microsoft Learn, Policy CSP Update (ConfigureDeadlineForQualityUpdates, ConfigureDeadlineGracePeriod): https://learn.microsoft.com/en-us/windows/client-management/mdm/policy-csp-update (read 19 Nov 2026) - Microsoft Learn, Manage device restarts after updates: https://learn.microsoft.com/en-us/windows/deployment/update/waas-restart (read 19 Nov 2026) - Microsoft Learn, Manage Windows quality updates in Intune (expedite requirements): https://learn.microsoft.com/en-us/intune/intune-service/protect/windows-10-expedite-updates (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/windows-endpoint-os/ --- # The Linux server and SSH playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** OpenSSH 10.6p2, released 5 Nov 2026 for CVE-2026-123158, closes a password bypass that needs six conditions at once, so most Linux servers are out of scope in minutes. The steps find the few that are in scope, give fixed Alpine and kernel package versions per distribution, and cover the reboot. CVE-2026-123158 is a login bypass in sshd, and it is narrow. It needs six conditions at once: sshd built against musl libc, no PAM, password authentication on, an account whose password field is *, another account with a SHA-256 or SHA-512 hash, and a reachable SSH port. An attacker sends a password of 257 to 1024 bytes for the starred account. musl's crypt() returns * when it fails, the stored field is also *, the two match, and the login succeeds. Debian, Ubuntu and Red Hat use glibc and PAM and are not affected. The hosts that are affected are Alpine, and the embedded images built the same way with OpenWrt, Buildroot or Yocto. What makes Linux servers different from the other playbooks is that the work splits in two. The SSH flaw is fixed by a package upgrade and a restart of sshd, and the upgrade alone does not replace the running daemon. The kernel flaws that CISA also lists for Linux are local privilege escalations that are fixed only by a new kernel and a reboot, and the distribution fix often shipped months before CISA listed the flaw. A shell gained through sshd and a local root flaw in the kernel are different bugs; no source reports them used together, but on one host the first hands an attacker a user and the second hands them root, so this page checks both. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Decide which hosts are in scope with one script, run as root on every host that runs sshd: b=$(command -v sshd); ldd "$b" 2>&1 \| grep -q musl && echo musl=yes \|\| echo musl=no; ldd "$b" 2>&1 \| grep -qi pam && echo pam=linked \|\| echo pam=none; sshd -T \| grep -iE '^(passwordauthentication\|usepam\|permitrootlogin)'; echo starred: $(awk -F: '$2=="*"{print $1}' /etc/shadow); echo hashed: $(awk -F: '$2 ~ /^\$[56]\$/{print $1}' /etc/shadow). A host is in scope when musl=yes, pam=none (or usepam no), passwordauthentication yes, and both account lists are non-empty. That is five of the six conditions; the sixth, a reachable port 22, is cut in step 3. Hosts with glibc are out of scope for this flaw. Also list embedded images from vendors, since Medcrypt reproduced the bypass on a stock Yocto image.
Five conditions have to meet on one host, and the script reads all five in a second, which turns "all Linux servers" into a short list. | Host, the five results, and the run time; the list of in-scope hosts goes to the platform team. | | 0 to 4 hours | Platform engineering | On each in-scope host that cannot be patched in the hour, turn password login off. First test key login in a second session and keep your first session open. Alpine's commands are: mkdir -p /etc/ssh/sshd_config.d; echo 'PasswordAuthentication no' > /etc/ssh/sshd_config.d/00-no-passwords.conf; sed -i.bak -e '1i PasswordAuthentication no' -e 's/^\(\s*PasswordAuthentication[^y#]*\)yes/\1no/i' /etc/ssh/sshd_config; sshd -t && rc-service sshd restart; sshd -T \| grep -i passwordauthentication. The last command must print passwordauthentication no. The drop-in file is read from Alpine 3.19 onward; the sed line covers older releases and Match blocks. To undo, delete the drop-in and copy sshd_config.bak back.
OpenSSH and Alpine both give this as the immediate step; it takes minutes and needs no package or change window. | The sshd -T output, the host and the time of the restart. | | 0 to 4 hours | Platform engineering | Cut port 22 to the bastion or VPN range. On AWS list the groups that open it to everyone with aws ec2 describe-security-groups --filters Name=ip-permission.from-port,Values=22 Name=ip-permission.cidr,Values=0.0.0.0/0 --query 'SecurityGroups[].[GroupId,GroupName]' --output text, then replace 0.0.0.0/0 with your bastion range. A rule that opens a range containing 22 needs a second query. On the host, ss -ltnp \| grep sshd shows what listens. Do this for in-scope hosts first, then for all.
The attacker needs a login prompt, and the in-scope list is only as good as the script, so closing the port also covers hosts the script missed. | Group IDs changed, old and new rule, and the time. | | 4 to 24 hours | Platform engineering | Patch. On Alpine run apk upgrade -U -a then rc-service sshd restart, and read the result with apk info -v openssh-server. Fixed package versions: edge 10.6_p2-r0; 3.24 10.3_p1-r2; 3.23 10.2_p1-r1; 3.22 10.0_p1-r11; 3.21 9.9_p2-r1; 3.20 9.7_p1-r6; 3.19 9.6_p1-r3; 3.18 9.3_p2-r4; 3.17 9.1_p1-r7; 3.16 9.0_p1-r6. Point releases 3.21.10, 3.22.8, 3.23.8 and 3.24.4 contain them; 3.16 to 3.20 have the packages but no new point release. Compare the package version, not the ssh -V banner, because Alpine's stable branches carry the fix under the old OpenSSH version. For OpenWrt, Buildroot and Yocto images, build OpenSSH 10.6p2 or apply commit 4353fdc43ef646c43a8dda7025cc1a898caf1482, which Medcrypt reports applies cleanly to OpenSSH 7.3p1 and later. Confirm the restart with ps -o pid,lstart,args -C sshd and compare the start time with the upgrade.
Alpine states that upgrading the package does not replace the running daemon, so a patched package with an old sshd process is still exposed. | Package version before and after, the sshd start time, and the host. | | 4 to 24 hours | Security lead | Look for logins that already happened on in-scope hosts. A starred account should only log in by key, so any password login for one is the bypass: for u in $(awk -F: '$2=="*"{print $1}' /etc/shadow); do grep "Accepted password for $u " /var/log/messages; done. Then look for GreyNoise's published traces: ls -l /etc/init.d/tac-agent /run/tac-agent.pid /root/.config/tac/agent.json.lock /tmp/py3.tar.gz; ls -d /opt/py3; grep -n vpnla /etc/hosts; sha256sum /etc/init.d/tac-agent against 7cf69bd6573a90d7bd287810b5c3746841cee6a7f77cadd016f34a97fc7b2361. Search logs for the addresses GreyNoise names, written here with dots in brackets: 183[.]220[.]91[.]15, 79[.]116[.]138[.]30, 20[.]255[.]63[.]133, 20[.]2[.]81[.]241, 209[.]248[.]50[.]17, 47[.]80[.]65[.]43, and 140[.]245[.]33[.]114 and 91[.]149[.]254[.]143 for a second cluster. Some may be compromised legitimate hosts. Review doas use, new users and authorized_keys. Alpine's rule: "If you cannot rule out a login, treat the machine as compromised."
CISA's KEV entry requires forensic triage, and GreyNoise says it cannot tell from the network whether a login used the bypass, so the host log is the only evidence. | The grep output, the hashes, and for each host a decision: clean, suspect or rebuild. | | 24 to 72 hours | Platform engineering | Rebuild images in the order given in step 6 of the [container base image playbook](/playbooks/container-base-image/); check an image with docker run --rm --entrypoint apk IMAGE info -v openssh-server and compare with the fixed list in step 4. For Alpine tiny-cloud images (qcow2, VHD and AMI) from 3.18.11 through 3.24.3, the alpine user has a * password and passwordless doas. They are safe on first boot and become exploitable as soon as any $6$ hash exists, for example when an administrator sets a console password or user-data adds a user with hashed_passwd. Images from 3.21.10, 3.22.8, 3.23.8, 3.24.4 and later carry the fix; for older images update openssh on every running instance.
A running instance created from an old image keeps the old package no matter what the repository now holds. | Image names and tags before and after, and the instances still on the old image. | | 24 to 72 hours | Security lead | Find the kernel exposure on all Linux servers, glibc ones included. Read the installed fix with rpm -q --changelog kernel \| grep -E 'CVE-2026-(46300\|53266)' on RPM systems and apt changelog linux-image-$(uname -r) \| grep -E 'CVE-2026-(46300\|53266)' on Debian and Ubuntu, because distributions backport the fix. Exposure: lsmod \| grep -E 'esp4\|esp6' for CVE-2026-46300, ebtables -t nat -L for a rule with -j snat and --snat-arp for CVE-2026-53266, and sysctl user.max_user_namespaces. Kernel floors that cover both flaws: RHEL 8 kernel-4.18.0-553.143.1.el8_10; RHEL 9 kernel-5.14.0-687.23.1.el9_8; RHEL 10 kernel-6.12.0-211.59.1.el10_2; Debian bullseye 5.10.259-1, bookworm 6.1.176-1, trixie 6.12.94-1; Ubuntu jammy 5.15.0-198.208, noble 6.8.0-146.146, resolute 7.0.0-31.31. Ubuntu focal 5.4.0-233.253 covers only CVE-2026-46300 and has no fix for CVE-2026-53266 yet, and Linux 5.4 is end-of-life. If you can reach a later kernel in the same reboot, CVE-2026-80521 (not in KEV; it affects RHEL 10, Debian bookworm and trixie, and Ubuntu noble, resolute and jammy HWE 6.8) needs RHEL 10 kernel-6.12.0-211.65.1.el10_2, Debian bookworm 6.1.190-1, trixie 6.12.111-1, Ubuntu noble 6.8.0-149.149, jammy HWE 6.8.0-149.149~22.04.1 and resolute 7.0.0-42.42.
CISA listed CVE-2026-53266 on 18 Sep 2026 and CVE-2026-46300 on 28 Oct 2026, with forensic triage required, 91 and 158 days after the stable kernels that fix them, so the first question is which hosts installed the fix and which of those have booted it. | Host, running kernel, package version of the installed kernel, and the exposure results. | | 24 to 72 hours | Platform engineering | Install the kernel and reboot, one node of each pair at a time. Check pending reboots with needs-restarting -r on RHEL and ls /var/run/reboot-required on Debian and Ubuntu, and compare uname -r with the newest installed kernel. Red Hat also ships kpatch updates for RHEL 8, 9 and 10 that avoid the reboot. Order: servers that run third-party code or give shell access first (shared shells, CI runners, hosts with web applications), then the rest.
A kernel package that is installed but not booted protects nothing, and the local flaws are used by an attacker who already has a shell. | Host, reboot time, and uname -r after the reboot. | | 24 to 72 hours | Security lead | Rescan with an authenticated scan and close with a list of exceptions. For each exception write the host, the owner, the date it will reach the fixed package or kernel, and the mitigation in place. The KEV due date for CVE-2026-123158 was 12 Nov 2026, 3 days after it was listed on 9 Nov; the two kernel flaws had due dates 3 days after listing too.
CISA's due date for these three Linux items is 3 days after listing, against 14 days for the Windows flaws, so the rescan is also the proof you hand over. | Scan date, the count in scope and fixed, and the exceptions list. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet For the SSH flaw the fix exists, so the question is images that cannot take 10.6p2 or hosts that cannot restart sshd. Turn off password login: OpenSSH's own wording for musl systems without PAM is to set PasswordAuthentication=no or disable SSH immediately. The cost of the first is that nobody can log in with a password, so keep console or key access ready; the cost of the second is all remote administration of that host. Medcrypt, a third party, reports that MaxAuthTries 1 blocks its reproduction and MaxAuthTries 2 does not; treat it as a stopgap, because a user who mistypes a password once is disconnected. Reading the preconditions, you can also make the host fall out of scope by removing every $5$ and $6$ password hash, at the cost of password logins at the console. Your own rebuild of an image with the fix commit is the lasting answer. For the kernel flaws, mitigations come from Red Hat. CVE-2026-46300: write install esp4 /bin/false and install esp6 /bin/false into /etc/modprobe.d/dirtyfrag.conf and run rmmod esp4 esp6; IPsec stops working, so check lsmod first. Or set user.max_user_namespaces=0 in /etc/sysctl.d/dirtyfrag.conf and run sysctl --system; rootless containers, sandboxed browsers and Flatpak can stop working. CVE-2026-53266: remove ebtables SNAT rules that rewrite ARP, which stops MAC translation for bridged virtual machines. Kernel 5.4 and Ubuntu focal have no fix for CVE-2026-53266, and CISA says the product may be end-of-life, so the answer there is an upgrade project. Take a host offline when a musl host with all six conditions faces the internet and neither patch nor password-off is possible, and for a kernel flaw only when untrusted users or code run on a host that fits none of the mitigations. ## CVE-2026-123158 worked through, hour by hour OpenSSH made the fix public and shipped 10.6p2 at 09:00 UTC on 5 Nov 2026, with Alpine's advisory and packages the same day. The sequence that followed is on GreyNoise's public timeline, and it shows how little time passed between the fix and the first successful attack. | When | What happened | Source | | --- | --- | --- | | 4 Nov 2026, 21:18 UTC | Damien Miller commits the OpenSSH fix, 4353fdc43ef6, with the message "avoid password authn bypass". It is not public yet. | [source](https://api.github.com/repos/openssh/openssh-portable/commits/4353fdc43ef646c43a8dda7025cc1a898caf1482) | | 5 Nov 2026, 09:00 UTC | The fix is pushed to OpenSSH's public repository and OpenSSH 10.6p2 is released at 09:01. Alpine's advisory, dated the same day, lists fixed packages for every release back to 3.16. | [source](https://www.openssh.org/releasenotes.html#10.6p2) | | 5 Nov 2026, 12:19 UTC | The researcher's write-up, which gives the exact trigger, appears 3 hours 18 minutes after the release. | [source](https://sshole.events.greynoise.io/) | | 5 Nov 2026, 13:16 UTC | Exploit code is published on GitHub, 2.5 hours before the CVE record is published. | [source](https://sshole.events.greynoise.io/) | | 5 Nov 2026, 15:46 UTC | MITRE publishes CVE-2026-123158. NVD lists it at 16:17 UTC. The CVSS base score of 8.1 comes from the MITRE CVE record; NVD has no rating of its own. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158) | | 7 Nov 2026, 00:46 UTC | GreyNoise records the first successful exploitation of one of its sensors, 39 hours 46 minutes after the fix went public. | [source](https://sshole.events.greynoise.io/) | | 7 Nov 2026, 10:12 UTC | In one intrusion GreyNoise tracks as GT10, the first command runs on a victim host (cat /etc/os-release), followed by a check for passwordless doas. At 10:54 the attacker pins its command domain in /etc/hosts and writes files. | [source](https://sshole.events.greynoise.io/) | | 9 Nov 2026 | CISA adds CVE-2026-123158 to KEV with forensicTriage Yes and a due date of 12 Nov 2026. | [source](https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json) | The 72-hour clock from the 09:00 fix on 5 Nov ends at 09:00 on 8 Nov, a day before CISA listed the flaw and 4 days before its due date. The first step takes minutes and the second under an hour, so a team that started at hour 0 had password login off at least 35 hours before the first successful attack at 00:46 on 7 Nov, even if the step took the full 4 hours. The patch and sshd restart fall inside hours 4 to 24, which still leaves 15 hours 46 minutes before that attack. The exposure windows were 39 hours 46 minutes from fix to first attack, and 32 hours 29 minutes from the NVD listing to it. A team that began at the KEV listing on 9 Nov began 2 days after the first attack. ## Related pages - [The AI infrastructure playbook for the first 72 hours](/playbooks/ai-infrastructure/) - [The Windows operating system playbook for the first 72 hours](/playbooks/windows-endpoint-os/) - [The developer collaboration platform playbook for the first 72 hours](/playbooks/developer-collaboration-platform/) ## Sources - Alpine Linux security advisory, 5 Nov 2026 (fixed packages, commands, detection): https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/ (read 19 Nov 2026) - OpenSSH 10.6p2 release notes: https://www.openssh.org/releasenotes.html#10.6p2 (read 19 Nov 2026) - OpenSSH fix commit 4353fdc43ef6: https://api.github.com/repos/openssh/openssh-portable/commits/4353fdc43ef646c43a8dda7025cc1a898caf1482 (read 19 Nov 2026) - GreyNoise, SSHole (CVE-2026-123158) activity timeline and attacker analysis: https://sshole.events.greynoise.io/ (read 19 Nov 2026) - Medcrypt, independent reproduction and mitigation guidance: https://www.medcrypt.com/blogs/asterisk-hole-cve-2026-123158-when-a-locked-account-is-the-way-in (read 19 Nov 2026) - NVD record for CVE-2026-123158: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026) - CISA Known Exploited Vulnerabilities feed, catalog version 2026.11.18 (CVE-2026-123158, CVE-2026-46300, CVE-2026-53266): https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026) - Linux 6.1.176 change log (fix for CVE-2026-53266, 19 Jun 2026): https://cdn.kernel.org/pub/linux/kernel/v6.x/ChangeLog-6.1.176 (read 19 Nov 2026) - Linux 6.1.174 change log (fix for CVE-2026-46300, 23 May 2026): https://cdn.kernel.org/pub/linux/kernel/v6.x/ChangeLog-6.1.174 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/linux-server-ssh/ --- # The AI infrastructure playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** LiteLLM, Langflow and MLflow servers are often started by one engineer, left open, and loaded with model-provider keys and cloud roles. CISA listed flaws in all three between 4 Aug and 2 Sep 2026. Find the servers, block the exposed routes, upgrade, then rotate every key the tool stored. Self-hosted AI gateways, workflow builders and experiment trackers behave like internal admin tools that were put on the internet for convenience. LiteLLM CVE-2026-59822 accepted a made-up Authorization header on its MCP endpoint and fell back to an empty auth object. Langflow CVE-2026-9198 let any network caller fetch a superuser token from an auto-login route and then run Python through a code-validation route. MLflow CVE-2026-64849 let an unauthenticated caller make the tracking server request internal addresses and read the answer. In all three the attacker needs no account, and the prize is what the server holds: provider API keys, stored secrets, tool access through MCP, and the cloud role of the host. Patching is awkward for three reasons. The tools are rarely in the asset inventory, because a data scientist ran pip install on a notebook host. The fixes ship long before the CVE: the LiteLLM fix was released on 14 May 2026, 55 days before the NVD record, and the Langflow fix on 23 Jun 2026, 24 days before. And the fixes change behavior: LiteLLM 1.84.0 is flagged by the vendor as a release with breaking changes, and Langflow's switch away from auto-login can lock you out of an existing install's flows. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Find the servers, including the unregistered ones. On hosts: pip list 2>/dev/null \| grep -Ei 'litellm\|langflow\|mlflow'. In containers: docker ps --format '{{.Image}} {{.Ports}}' \| grep -Ei 'litellm\|langflow\|mlflow'. In Kubernetes: kubectl get deploy,sts -A -o wide \| grep -Ei 'litellm\|langflow\|mlflow'. From outside, scan your own public ranges for the ports these tools commonly use: Langflow's documentation uses 7860, LiteLLM's proxy commonly runs on 4000, MLflow's server on 5000. Ask the data and ML teams for the ones they started themselves.
These servers are not on the asset list, and the hour spent finding them is the hour the exposed ones stay open. | Per server: host or pod, tool, version, port, who started it, internet-reachable yes or no, and which cloud role it runs under. | | 0 to 4 hours | Platform engineering | Check each server against the affected range and the precondition. LiteLLM below 1.84.0: pip show litellm, then grep -n -A6 mcp_servers config.yaml and grep -n -B2 -A8 allow_all_keys config.yaml; with no mcp_servers block there are no MCP tools to reach. Langflow 1.0.0 through 1.10.0: pip show langflow, then env \| grep LANGFLOW_AUTO_LOGIN and the .env file; treat unset on 1.10.0 or earlier as enabled. MLflow 3.3.0 up to but excluding 3.15.0: pip show mlflow, then ps -ef \| grep 'mlflow server' and read --backend-store-uri; webhooks need a SQL store, and the default SQLite store counts.
The version alone does not say whether you are exposed; the configuration decides, and it tells you which block to apply first. | Per server: version, the three precondition answers (config item present, auth state, store type) and exposed yes or no. | | 0 to 4 hours | Platform engineering | Cut the reachable path now. LiteLLM: disable the MCP routes, or block /mcp/ and the related MCP paths (for example /mcp-rest/) at the reverse proxy, then check with curl -s -o /dev/null -w '%{http_code}\n' https://PROXY_HOST/mcp/ from outside, which must time out or be blocked. Langflow: put an authenticating reverse proxy in front, and set LANGFLOW_AUTO_LOGIN=false only after LANGFLOW_SUPERUSER_PASSWORD is set, or you lose access to the superuser's flows. MLflow: load the auth plugin and keep the server off untrusted networks. For all three, remove the public listener and allow only the networks that need it.
A route block takes minutes and stops the unauthenticated call, while the upgrade needs a test cycle. | The proxy rule or setting changed, the test result from outside, and the time it took effect. | | 4 to 24 hours | Platform engineering | Upgrade: LiteLLM to 1.84.0 or later, after reading the release notes for breaking changes and testing clients in staging; Langflow to 1.10.1 or later, which stops issuing the 365-day superuser token and no longer executes code in validate_code; MLflow to 3.15.0 or later, and to 3.16.0 or later on IPv6 or NAT64 networks, because 3.15.0 left four IPv6-transition bypasses. Pin the version in the image tag or requirements file so a rebuild cannot go back.
The route block is a stopgap that breaks MCP clients; the upgrade is the actual fix, and IBM and the MLflow advisory list no workaround for Langflow or MLflow. | Version before and after, the pinned reference, and the staging test result for LiteLLM. | | 4 to 24 hours | Security lead | Rotate everything the server stored. List the secrets: LiteLLM provider keys and issued virtual keys; Langflow global variables, API keys and every session issued under auto-login, including the 365-day superuser token named in the 1.10.1 release notes; MLflow artifact-store and database credentials. Turning auto-login off does not revoke earlier sessions or API keys, so revoke them by hand. Then ask each provider for its usage log from the exposure window and compare it with what you expect.
Whoever read the stored keys can use them from anywhere, and the patch does nothing about keys already copied. | Each secret rotated, with the time, the provider and the usage-log result. | | 4 to 24 hours | Platform engineering | For MLflow, close the metadata path the SSRF reaches. From the server host run curl -s -m 3 -o /dev/null -w '%{http_code}\n' http://169.254.169.254/ to see whether it answers. On AWS require IMDSv2 and keep containers out: aws ec2 modify-instance-metadata-options --instance-id --http-tokens required --http-put-response-hop-limit 1. Check the current setting with aws ec2 describe-instances --instance-ids --query 'Reservations[].Instances[].MetadataOptions.HttpTokens'. Software that still uses IMDSv1 on the host will break, so list it first.
The MLflow bug returns response bodies from internal addresses, and metadata credentials are the most valuable thing there. | The metadata setting before and after, the software that broke, and the fix applied. | | 4 to 24 hours | Security lead | Hunt in the proxy logs. Langflow: a request to /api/v1/auto_login followed by one to /api/v1/validate/code from the same source or from outside your network. LiteLLM: requests to /mcp/ or /mcp-rest/ from unknown addresses carrying an Authorization value that matches no key you issued; Wiz saw probes using bearer tokens of one character against model-listing endpoints. MLflow: POST to /api/2.0/mlflow/webhooks and /api/2.0/mlflow/webhooks/{id}/test, and a list of registered webhooks pointing at hosts you do not recognize. On the network: connections from the server to 169.254.169.254 or private ranges it has no reason to contact.
None of the three vendors published a log signature, so the route names and the odd bearer value are the only search terms there are. | The search text, the log range from the earliest affected release on the host, and each hit with source address and time. | | 24 to 72 hours | Security lead | Look on the host. Run ps -eo pid,ppid,user,cmd --forest and look for the AI server process with a shell or an unfamiliar child. Check temporary paths for what Wiz found on its honeypots, which it did not tie to these CVEs: /tmp/x86_64 or /tmp/amd64, /tmp/.dbus-cache/gmon, a miner renamed unicorn under /app/data/.claude/, staging address 94.26.106[.]29 and 185.62.1[.]8, and the mining domain pool.hashvault[.]pro. For Langflow, CISA marks forensic triage as required, so copy the logs and take a memory image before you clean up.
A cryptominer is the usual payload on these servers, and it shows up as a process before it shows up as a bill. | The process list, any file found with its hash and timestamp, and the decision to clean or rebuild the host. | | 24 to 72 hours | Cloud engineer | Shrink what the server's role can do. List the attached policies with aws iam list-attached-role-policies --role-name and remove wildcard actions and any access to buckets or models the tool does not use. Deny outbound traffic from the host except to the model providers and stores it needs.
The next flaw in the tool will be exploited the same way, and a narrow role limits what it can reach. | The role before and after, and the egress rule set. | | 24 to 72 hours | Application owner | Register each server with an owner and a release feed. Subscribe to the project's GitHub releases and treat any release note about auth, validation or code execution as a security fix. Put the server behind your single sign-on, then run the first check from the 0 to 4 hour window again to prove the route block can be lifted.
The fixes arrive weeks before the CVE, so the people who watch releases patch before the exploitation starts. | Owner name, the feed subscribed, the next review date. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet Vendor workarounds in this class are rare. LiteLLM published one (disable MCP routes or block /mcp/ at the gateway). IBM lists no workaround for Langflow, and the MLflow advisory lists none. The controls that exist are the ones in the route-block step: an authenticating proxy, auth turned on, removing the listener from untrusted networks, IMDSv2 and egress limits. Each one costs: MCP clients and agents stop working behind a block, users sign in again after auth is on, and software on a host that uses IMDSv1 breaks. Decide on offline by what the server holds and who can reach it. If the server is reachable from the internet, holds provider keys or a broad cloud role, and no authenticating proxy can go in front of it inside four hours, stop it. CISA's required action for the AI flaws it listed tells you to discontinue use of the product if mitigations are unavailable. Stopping a LiteLLM gateway or a Langflow editor costs the applications behind it, so name those before the decision, and record who made it and when. ## CVE-2026-9198 worked through, hour by hour Langflow OSS 1.0.0 through 1.10.0 issued a superuser token to any caller of the auto-login route, and a code-validation route then executed Python at function-definition time. CVSS base score 9.8. The dated events, from the release notes, IBM, NVD and CISA, run for 42 days before the KEV listing. The sources give dates, not hours, so the table keeps days. | When | What happened | Source | | --- | --- | --- | | 23 Jun 2026, Day 0 | Langflow 1.10.1 is released. The notes list 'do not execute code in validate_code' and 'stop issuing 365-day superuser token via auto_login', without naming a CVE. | [source](https://github.com/langflow-ai/langflow/releases/tag/v1.10.1) | | 2 Jul 2026, Day 9 | IBM publishes its security bulletin, tells users to upgrade Langflow OSS to 1.10.1 and lists no workaround or mitigation. | [source](https://www.ibm.com/support/pages/node/7278927) | | 17 Jul 2026, Day 24 | NVD publishes CVE-2026-9198. The CISA coordinator SSVC entry stamped the same day lists exploitation as active and the flaw as automatable. | [source](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-9198) | | 4 Aug 2026, Day 42 | CISA adds the CVE to KEV with forensic triage required and a due date of 7 Aug 2026. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-9198) | | 7 Aug 2026, Day 45 | KEV due date, three days after the listing. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-9198) | | 27 Aug 2026, Day 65 | Wiz publishes 90 days of honeypot telemetry, including attackers who compromised a Langflow honeypot and dropped cryptominers. Wiz does not tie those attacks to this CVE. | [source](https://www.wiz.io/blog/ai-infrastructure-honeypot) | The playbook turns the 42 days into three. An operator who followed the release feed in the last step upgraded on 23 Jun and never saw the flaw. One who started the 72-hour playbook on the IBM bulletin of 2 Jul finished by 5 Jul, 12 days before the SSVC entry marked exploitation active and 30 days before KEV. One who waited for the KEV listing had 3 days to upgrade, rotate the 365-day superuser token and search for the auto_login and validate/code sequence, with forensic triage required. The same pattern held for the other two tools: LiteLLM's fix (v1.84.0, 14 May 2026) was out 111 days before its KEV listing on 2 Sep 2026, and MLflow 3.15.0 (31 Jul 2026) 19 days before its KEV listing on 19 Aug 2026. ## Related pages - [The developer collaboration platform playbook for the first 72 hours](/playbooks/developer-collaboration-platform/) - [The Linux server and SSH playbook for the first 72 hours](/playbooks/linux-server-ssh/) - [Application runtime playbook: the first 72 hours when a flaw is in PHP or another language runtime on your servers](/playbooks/application-runtime/) ## Sources - LiteLLM advisory GHSA-7488-6r32-c95q: https://github.com/BerriAI/litellm/security/advisories/GHSA-7488-6r32-c95q (read 19 Nov 2026) - Wiz on AI infrastructure honeypot: https://www.wiz.io/blog/ai-infrastructure-honeypot (read 19 Nov 2026) - IBM security bulletin for Langflow: https://www.ibm.com/support/pages/node/7278927 (read 19 Nov 2026) - Langflow 1.10.1 release notes: https://github.com/langflow-ai/langflow/releases/tag/v1.10.1 (read 19 Nov 2026) - MLflow advisory GHSA-7gwp-5pfp-969j: https://github.com/mlflow/mlflow/security/advisories/GHSA-7gwp-5pfp-969j (read 19 Nov 2026) - LiteLLM v1.84.0 release: https://github.com/BerriAI/litellm/releases/tag/v1.84.0 (read 19 Nov 2026) - MLflow v3.15.0 release: https://github.com/mlflow/mlflow/releases/tag/v3.15.0 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-9198: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-9198 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-59822: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-59822 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-64849: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-64849 (read 19 Nov 2026) - NVD record for CVE-2026-9198: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-9198 (read 19 Nov 2026) - NVD record for CVE-2026-59822: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-59822 (read 19 Nov 2026) - NVD record for CVE-2026-64849: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-64849 (read 19 Nov 2026) - Langflow documentation: API keys and authentication: https://docs.langflow.org/api-keys-and-authentication (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/ai-infrastructure/ --- # The developer collaboration platform playbook for the first 72 hours Playbook, sources read 19 Nov 2026 **The short answer.** Self-managed Atlassian products hold source code, tickets and build plans, and CVE-2026-21589 lets an unauthenticated caller read files in the web root of eight of them. Atlassian published fixed versions on 5 Oct 2026 and CISA listed the flaw on 15 Oct. Block the traversal pattern first, then upgrade each product. Bitbucket, Confluence, Jira Software, Jira Service Management, Bamboo, Crowd, Crucible and Fisheye are rarely patched together. Different teams own them, they run on different release lines, and Crowd sits behind most of the others as the user directory. A single flaw in the shared web layer hits all of them at once. CVE-2026-21589 is an arbitrary file read: an unauthenticated caller who knows the exact name and path of a file inside the web application root can fetch it. It does not list directories, so the damage depends on what sits under the root, such as WEB-INF configuration and, for Crowd, the application password in crowd.properties. What makes it hard to patch is the spread. Atlassian says every version before the fixed builds is affected, so an instance on a release line that has no fixed build must move to a listed line. Atlassian also says it cannot confirm whether any given instance was affected, and tells you to check all of them for signs of compromise. Fixes ship as maintenance releases, not binary patches, so the upgrade is a normal upgrade with a normal test cycle. The first job is to stop the reachable path, not to start that cycle. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | List every instance of the eight products and read each version from the administration System information page. Match it to the fixed builds: Bitbucket Data Center 9.4.26, 10.2.8 or 10.5.1; Confluence Data Center 9.2.26 or 10.2.19; Jira Service Management Data Center 5.12.40, 10.3.26 or 11.3.12; Jira Software Data Center 9.12.40, 10.3.26 or 11.3.12; Bamboo Data Center 10.2.24 or 12.1.12; Crowd Data Center 6.3.7, 7.0.3, 7.1.7 or 7.2.4; Crucible 4.9.15; Fisheye 4.9.15. Take the fixed build on your own line. If your line is not listed, it is end of life: Atlassian publishes no per-line mapping and says to patch to "the fixed LTS version or later", so move to the oldest listed build for the product (Bitbucket 9.4.26, Confluence 9.2.26, Jira Service Management 5.12.40, Jira Software 9.12.40, Bamboo 10.2.24, Crowd 6.3.7, Crucible and Fisheye 4.9.15) and read Atlassian's upgrade notes for whether your version can go straight there. Atlassian Cloud sites were patched by Atlassian and need no action.
All eight products share the flaw but not the fixed version, and Crucible, Fisheye and Bamboo are the ones that miss an inventory. | One row per instance: product, version, fixed build for its line, internet-reachable yes or no, owner. Mark any end-of-life line. | | 0 to 4 hours | Platform engineering | Take instances reachable from the internet off it, or restrict them to the corporate network or VPN. Atlassian says this applies even to instances with user authentication. In the reverse proxy or firewall, allow the corporate and VPN ranges and deny the rest. Include Bitbucket mirrors and mirror farm nodes, which serve the same web layer.
The attacker needs no account, so only network position separates a reachable instance from a readable one. | The rule text, the time it went live, and the list of integrations that broke (webhooks from outside, build agents, mobile clients). | | 0 to 4 hours | Platform engineering | Add Atlassian's Option 1 rule at the WAF or proxy for every product: block any URL that matches (?is).*(?:/\|\\\|::\|%(?:25)*(?:2f\|5c)\|(?::\|%(?:25)*3a){2})(?:\.\|%(?:25)*2e){2}(?:/\|\\\|::\|%(?:25)*(?:2f\|5c)\|(?::\|%(?:25)*3a){2}\|;\|%(?:25)*3b\|$).* , which means two dots next to a slash, backslash or double colon in plain, URL-encoded and double-encoded form. Then test it from outside with curl --path-as-is -s -o /dev/null -w '%{http_code}\n' against your own host using /x/%2e%2e/y, /x/%252e%252e%5cy and /x/..;/y. All three must be blocked. A path like /x/file..txt must still work.
The rule protects the instances you cannot patch today, and a rule that misses the double-encoded form protects nothing. | The three test requests with the response code of each, the rule version, and the products covered. | | 0 to 4 hours | Security lead | Preserve and search the access logs back to 5 Oct 2026. Copy the logs of the proxy and each product, then URL-decode each request line up to two passes and search for two dots directly next to a slash, backslash or double colon. A quick pass on raw logs: grep -Ei '(/\|\\\|::\|%(25)*(2f\|5c))(\.\|%(25)*2e){2}(/\|\\\|::\|%(25)*(2f\|5c)\|;\|%(25)*3b\|[ ?]\|$)' access*.log. Then sort the hits by response size and status. A 200 with a body, on a path under WEB-INF, means that file was read. The vendor regex also catches the form watchTowr described, in which a request to the plugin resource path (the segment named download/resources) uses double colons as parent-directory segments and names a file under WEB-INF. Search separately for any request URL that contains both the text WEB-INF and a double colon: grep -Ei 'WEB-INF' access*.log \| grep -E '::\|%3a%3a' (run it on the decoded log too). Search separately for requests to the Crowd REST paths /crowd/rest/usermanagement/1/ from addresses that are not your own applications, since a stolen application password is used there.
Atlassian cannot tell you whether an instance was hit, and a 200 on a traversal request is the only direct evidence you will get. | The log range, the decode method, each hit with address, time, path and response code, and the list of files that were returned. | | 4 to 24 hours | Platform engineering | Upgrade in this order: Crowd first, because the directory is read by the others and its crowd.properties holds an application password; then the products with internet-reachable instances; then the rest. Back up, shut down each node, upgrade to the fixed build, restart, and confirm the version on the System information page. Do not skip the mirrors: upgrade every Bitbucket mirror and mirror farm node.
Crowd's credentials are the most valuable file under any web root, and an unpatched mirror serves the flaw as well as the primary. | Version before and after for each node, the backup location, the start and end time of the outage. | | 4 to 24 hours | Platform engineering | If an instance cannot be upgraded within 24 hours, apply a rewrite rule at the application. Confluence, Jira Service Management, Jira Software, Bamboo and Crowd (Option 2): back up, shut the node down, add inside the application's Context element in conf/server.xml (for Crowd, apache-tomcat/conf/Catalina/localhost/crowd.xml or apache-tomcat/conf/server.xml, whichever holds the Context), then put Atlassian's rewrite.config in confluence/WEB-INF, atlassian-jira/WEB-INF, bamboo/WEB-INF or atlassian-bamboo/WEB-INF, or crowd-webapp/WEB-INF, and restart. Each rule pairs a RewriteCond using the regex against %{REQUEST_PATH} (and a second against %{REQUEST_URI} as a raw fallback) with RewriteRule (?s)^.*$ /?denied [F,L]. Bitbucket (Option 3): back up, add a new first rule to app/WEB-INF/urlrewrite.xml on every node, mirror and mirror farm node, with the regex in its from element, status 404 and target /mvc/error404, then restart.
A host-level rule keeps working when traffic bypasses the WAF, for example from inside the corporate network. | The file changed, the diff, the restart time per node and a negative test result for each. | | 4 to 24 hours | Security lead | List what a reader could take. On each node list the web root (ls -la confluence/WEB-INF, atlassian-jira/WEB-INF, bamboo/WEB-INF, crowd-webapp/WEB-INF, app/WEB-INF) and open any configuration file for credentials. For Crowd, check crowd.properties. Rotate every credential found in a file under the web root if the log search shows a request for it, and treat it as exposed anyway on any instance that was internet-reachable while unpatched. Files outside the web application root are not covered by this flaw.
The attacker needs the exact path, so the exposure is bounded by the files under the root, and that list is short enough to review in full. | Each file with a credential, whether it was requested in the logs, the credential rotated and the time. | | 24 to 72 hours | Application owner | Test what depends on each product before restoring access: application links between Jira, Confluence and Bitbucket, Bamboo agents and their links to Bitbucket, Crowd-backed logins, webhooks and Marketplace apps. Check that the Option 1, 2 or 3 rules did not block a legitimate URL; the vendor warns any URL containing such a sequence is denied. Then restore internet access only through the intended gateway.
Restoring exposure before the dependants are tested trades a security incident for an outage. | A pass or fail per integration, the time of restore, and who approved it. | | 24 to 72 hours | Security lead | Close or escalate the compromise question using the evidence: no hits and no exposed credentials means close with the log range noted; hits that returned content, or a CISA forensic triage requirement you are subject to, mean preserve disk images and logs of the node first, then start incident response. Keep the Option 1 rule in place after the upgrade until your next review.
CISA's entry requires forensic triage, and a decision made without writing down the evidence cannot be defended later. | The decision, the evidence behind it, who decided and the time. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet Two situations leave you without a fix. The first is an end-of-life release line: Atlassian says all versions are affected and recommends the fixed LTS version or later, so a line with no fixed build must be upgraded to a listed line, and until then the WAF rule, the Tomcat rewrite rule or the Bitbucket urlrewrite rule is the only protection. The second is a fixed build you cannot test inside 24 hours. Both cost you something: the rewrite options need a restart on every node and deny any URL with the sequence, and the WAF rule is only as good as your test of the encoded forms. Take the instance off the internet at once if it is reachable, and keep it off until it is patched or protected by a tested rewrite rule. The flaw gives read access to files with a known path, so the decision turns on what is under the web root: if the root holds credentials you cannot rotate quickly, or Crowd is exposed, take the service down rather than leave it running behind a rule you have not proven. Atlassian itself advises removing instances from the internet until you can patch or apply mitigations. ## CVE-2026-21589 worked through, day by day Atlassian published the advisory on 5 Oct 2026 for Bitbucket, Confluence, Jira Service Management, Jira Software, Bamboo and Crowd Data Center, plus Crucible and Fisheye. CVSS 4.0 base score 9.3. CISA lists the flaw in KEV, but neither Atlassian nor CISA publishes a date of first exploitation. The dated events are the advisory, a research post and the CISA listing, and they are given by day. | When | What happened | Source | | --- | --- | --- | | Before 5 Oct 2026 | Atlassian has patched its Cloud products and reports that its investigation found no evidence of exploitation there. | [source](https://confluence.atlassian.com/security/cve-2026-21589-arbitrary-file-access-vulnerability-impacts-multiple-products-1870495748.html) | | 5 Oct 2026, Day 0 | Atlassian publishes the advisory with fixed versions for all eight products and three mitigation options. NVD publishes the record the same day. | [source](https://confluence.atlassian.com/security/cve-2026-21589-arbitrary-file-access-vulnerability-impacts-multiple-products-1870495748.html) | | 6 Oct 2026, Day 1 | watchTowr Labs publishes its analysis. The cause is the web resource handler in atlassian-plugins-webresource (6.0.7 vulnerable, 6.0.8 changed), which turns a double colon in a request path into a slash. A request to the plugin resource path of Jira, built with double colons as parent-directory segments, returns WEB-INF/web.xml. | [source](https://labs.watchtowr.com/you-wont-hear-about-these-even-in-myths-atlassian-jira-confluence-and-more-pre-auth-arbitrary-file-read-cve-2026-21589/) | | 6 Oct 2026, Day 1 | The same post gives the request forms for the other products: Confluence uses its plugin resource path under the /s/ static prefix, and Bitbucket does the same with its server plugin, naming a different WEB-INF file because web.xml is blocked there. The researchers read files in the web application root only; they could not leave the Tomcat context. | [source](https://labs.watchtowr.com/you-wont-hear-about-these-even-in-myths-atlassian-jira-confluence-and-more-pre-auth-arbitrary-file-read-cve-2026-21589/) | | 6 Oct 2026, Day 1 | watchTowr shows what makes the read serious: Jira's WEB-INF/classes/crowd.properties holds application.name and, in plaintext, application.password for the Crowd directory. With them a caller reaches Crowd's REST API (its user search and user creation endpoints under /crowd/rest/usermanagement/1/) and can list users or create one. They note that a Crowd IP allow list makes this harder. | [source](https://labs.watchtowr.com/you-wont-hear-about-these-even-in-myths-atlassian-jira-confluence-and-more-pre-auth-arbitrary-file-read-cve-2026-21589/) | | 15 Oct 2026, Day 10 | CISA adds the CVE to KEV with forensic triage required and a due date of 18 Oct 2026. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-21589) | The playbook's steps all fall in the first three days of that table. KEV came on day 10, so every step in the first 72 hours was taken seven days before CISA's listing and ten days before its due date. An organization that waited for the listing had three days to patch eight products on several release lines, and no early log search to show whether anyone had read a file in the ten days before. Neither Atlassian nor CISA gives an exploitation date, so that log search back to 5 Oct is the only way to date a read. ## Related pages - [Application runtime playbook: the first 72 hours when a flaw is in PHP or another language runtime on your servers](/playbooks/application-runtime/) - [The AI infrastructure playbook for the first 72 hours](/playbooks/ai-infrastructure/) - [The communications playbook for the first 72 hours](/playbooks/communications/) ## Sources - Atlassian advisory CVE-2026-21589: https://confluence.atlassian.com/security/cve-2026-21589-arbitrary-file-access-vulnerability-impacts-multiple-products-1870495748.html (read 19 Nov 2026) - watchTowr Labs on CVE-2026-21589: https://labs.watchtowr.com/you-wont-hear-about-these-even-in-myths-atlassian-jira-confluence-and-more-pre-auth-arbitrary-file-read-cve-2026-21589/ (read 19 Nov 2026) - CISA KEV entry for CVE-2026-21589: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-21589 (read 19 Nov 2026) - NVD record for CVE-2026-21589: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-21589 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/developer-collaboration-platform/ --- # Application runtime playbook: the first 72 hours when a flaw is in PHP or another language runtime on your servers Playbook, sources read 19 Nov 2026 **The short answer.** A flaw in the language runtime that serves your web requests is patched on the host: the package, the server module or pool, and its configuration. This playbook gives the PHP commands for hours 0 to 72, from finding which server API runs to the audit rules, worked through PHP2Shell. A runtime is the program that executes your application code, and a web-facing flaw in it is reached by an ordinary HTTP request before any of your code runs. For PHP2Shell, CVE-2026-109618 is triggered in multipart form parsing. With Apache mod_php that parsing happens before application code, so no login and no application bug is needed. The attacker gains code execution as the web worker account, with the application's files, database credentials and network position. PHP rates mod_php the highest risk. The runtime is hard to patch for reasons that are not about code. One host can run three PHP builds at once, such as mod_php, PHP-FPM and a command-line binary, and the command-line version proves nothing about the web one. Distribution packages fix the flaw under an old version number. The behavior depends on configuration that lives in php.ini, pool files, virtual hosts and .htaccess files. And a runtime upgrade touches every request the host serves. This playbook is about the PHP process and its configuration on servers and virtual machines. Images built from a base are handled by the container playbook. ## The steps, hour by hour | Window | Owner | Step, and why it comes here | Record to keep | | --- | --- | --- | --- | | 0 to 1 hour | Security lead | Find every web-serving PHP, by server API (SAPI), not by version alone. ps -ef \| grep -Ei 'php-fpm\|php-cgi\|frankenphp\|lsphp\|httpd\|apache2' \| grep -v grep lists running servers, and apachectl -M \| grep -i php (apache2ctl -M on Debian and Ubuntu) shows whether Apache loads mod_php. List packages: dpkg -l 'php*' 'libapache2-mod-php*' or rpm -qa 'php*'. A host that runs PHP only from the command line, for cron jobs, drops out of web scope. Mark mod_php first, because PHP rates it highest.
CISA's alert says the flaws affect every PHP serving web requests, and the order of work is set by the SAPI. | One row per host: the SAPI, the web server, the package name, and the owning team. | | 0 to 4 hours | Platform engineering | Resolve the version question per host. Run the web binary: php-fpm -v, php-cgi -v, or for mod_php php -v after confirming the module. Then ask the package for the fix, because distributions backport without changing the number: rpm -q --changelog php \| grep -i 109618 or apt changelog php8.3 \| grep -i 109618, and the same with 109619. Fixed upstream releases are 8.2.35, 8.3.36, 8.4.27 and 8.5.12. Anything on 8.1 or older has no upstream fix; mark it for step 10.
A version string below the fixed release can still be patched, and a number above it can hide a different build. | Host, binary run, version printed, changelog hit yes or no, and branch (8.5, 8.4, 8.3, 8.2, 8.1 or older). | | 0 to 4 hours | Security lead | Order exposure by what an anonymous client can do. From outside the network: curl -s -o /dev/null -w '%{http_code}\n' -X POST -F a=b https://host/index.php; any PHP response, even an error page, means the server passed a multipart POST to PHP. Find the handler paths in config: grep -R 'fastcgi_pass\\|SetHandler\\|proxy:fcgi\\|AddHandler' /etc/nginx /etc/apache2 /etc/httpd. A path that requires login before PHP removes the unauthenticated condition. Internet-facing mod_php with a public entry point goes first, FPM behind a login last.
The flaw is unauthenticated remote code execution, so who can send the POST decides the order. | Per vhost: reachable from internet, passes to PHP, requires login, SAPI. | | 0 to 4 hours | Platform engineering | On hosts that cannot be patched in hours, apply the stated interim settings as admin values and restart every worker. CVE-2026-109618: memory_limit=-1. In an FPM pool: php_admin_value[memory_limit] = -1. In Apache: php_admin_value memory_limit -1. CVE-2026-109619: php_admin_value[session.upload_progress.enabled] = 0, and on PHP 8.5 php_admin_value[fatal_error_backtraces] = 0. Restart: systemctl restart php8.3-fpm or apachectl graceful (restart the service if graceful does not take it). Prove the pool value with php-fpm -tt \| grep -i memory_limit, and check nothing resets it in .htaccess or a pool: grep -Rn 'memory_limit' /etc/php* /etc/php-fpm.d /etc/apache2 /etc/httpd. Costs: a runaway script can use all host memory, and upload progress stops. This is partial: PHP says a failed mmap() can still reach the vulnerable path, and only the upgrade is complete.
These settings narrow the demonstrated paths while the upgrade is scheduled, and admin values stop an application or .htaccess from undoing them. | The config change with its commit, the php-fpm -tt output, the restart time, and the exception end date. | | 4 to 24 hours | Platform engineering | Frontend limits, where the application allows them. Keep post_max_size conservative and enforce the decoded size at the frontend; lower max_input_nesting_level; bound input variables and multipart parts. If the application has no uploads or multipart forms, refuse them before PHP: in nginx, if ($content_type ~* "^multipart/form-data") { return 415; } in the server block, then repeat the step 3 curl and expect 415. Cost: file uploads and multipart forms stop working. Do not use RLIMIT_AS or RLIMIT_DATA as a memory cap: they make mmap() fail and trigger the vulnerable path.
A refusal before PHP runs removes the confirmed trigger for hosts with no upload use, with no change to the runtime. | The frontend rule, the test request and status code, and the list of routes that need uploads. | | 4 to 24 hours | Platform engineering | Upgrade the package and restart. Debian and Ubuntu: apt-get update && apt-get install --only-upgrade 'php8.3*' (use your branch). RHEL family: dnf upgrade 'php*'. Then restart the FPM service or Apache; a package install does not replace the running workers. Proof: ls -l /proc/$(pgrep -o php-fpm)/exe shows (deleted) when a worker still runs the old binary, and ps -o lstart= -p $(pgrep -o php-fpm) must be later than rpm -q --qf '%{INSTALLTIME:date}\n' php-fpm. Repeat the changelog grep from step 2 and the version commands. Run it across hosts with ansible web -m command -a 'php-fpm -v'.
Upgrading the package without restarting leaves the vulnerable code in memory, and the exe check catches it. | The package version after, the worker start time, the changelog hit, and the ansible output saved with the date. | | 4 to 24 hours | Application owner | Test on staging with real traffic patterns. The fix is a patch release within a branch, but it changes how out-of-memory is handled: PHP now soft-fails at the next VM safe point and hard-exits with _exit() at a second, higher memory threshold. Run composer check-platform-reqs, diff php -m against php-fpm -m before and after, upload the largest file your users send, run an upload at the post_max_size limit, and check session upload progress if the application uses it. Look for workers that exit with a memory limit error where the old build let a script run on.
The owner finds regressions on one host before the platform team rolls the change to all, and a failed upgrade is what pushes teams back to the interim setting. | The test list, the pass or fail, the largest upload tested, and the approval to roll out. | | 24 to 72 hours | Platform engineering | Remove the conditions CVE-2026-109619 needs, on patched and unpatched hosts. Find per-request overrides: grep -Rn 'php_value\\|php_flag\\|PHP_VALUE' /etc/apache2 /etc/nginx /etc/httpd and find /var/www -name .htaccess \| xargs grep -n 'php_'. Move them into php.ini, pool or server configuration applied at startup, because INI warnings raised during request startup are a confirmed consumer. Audit in isolation with log_errors=On, a working error_log and error_reporting=E_ALL in startup configuration, and review Apache, FPM and proxy logs for Deprecated and Warning lines before target scripts run, and for the string Maximum execution time exceeded. A clean log is not proof of safety, per PHP. Keep session.upload_progress.enabled=0 after patching if no application uses it.
The upgrade removes the flaw, but the audit shows which settings generated the warnings that made it reachable, and any later regression of the same kind. | The overrides found and moved, the audit period, the log lines matched, and the owner for each application setting. | | 24 to 72 hours | Security lead | Hunt for exploitation on the origin. Success is invisible at the edge. Search Apache and FPM error logs for worker crashes: grep -h 'AH00051' /var/log/apache2/error.log* (path varies by distribution). Look for web workers that launched shells: pstree -a -p $(pgrep -o php-fpm) \| grep -Ei 'sh\|bash\|curl\|wget', and unexpected outbound connections: ss -tnp \| grep -E 'php\|httpd\|apache2'. Start a recorder for the next days: auditctl -a always,exit -F arch=b64 -S execve -F uid=www-data -k webexec (replace www-data with your worker account), then ausearch -k webexec. In access logs, search the User-Agent php2shell-109618 and the multipart boundary B7kQpoc618 where bodies are logged, but treat both as stale: Fastly says user agents, source IPs and body sizes go stale within a day. Rule out React2Shell look-alikes: they carry a Next-Action header, which PHP traffic never does. CISA published a SIGMA rule. If a vulnerable host saw it, treat the worker account's credentials as exposed.
The exploit leaves no payload string to match, so the signal is a crash or a shell child process, and these searches are cheap after the fix is in. | The searches with time ranges and counts, the audit rule, and the hosts moved to incident response. | | 24 to 72 hours | Platform engineering | Handle runtimes with no upstream fix. For PHP 8.1 and older, ask the distributor for a backport and prove it with the changelog grep, or schedule a move to a supported branch. Until then, CISA tells operators to take PHP-based systems offline, and a WAF may give partial protection. Put each exception in the record with an owner and an end date.
These hosts are exploitable as shipped, and the longer they run the more the exception is a risk decision someone must own. | The exception list: host, branch, distributor answer, compensating control, owner, end date. | Print this table as the checklist: write the time each step was done in the record column. ## When no fix exists yet The runtime case without a fix is PHP 8.1 and older, and distributions that have not yet shipped a backport. For CVE-2026-109618 the interim controls are memory_limit=-1, conservative post_max_size with the decoded size enforced at the frontend, lower max_input_nesting_level, a multipart block at the frontend where uploads are not used, and a cgroup memory.max limit with the caveat that killing the whole cgroup can take down the FPM master and all workers. For CVE-2026-109619 there is no setting that covers every producer and consumer: fatal_error_backtraces=0 on 8.5, session.upload_progress.enabled=0, startup-only INI, and pm.max_requests=1, which makes each worker serve one request at a heavy throughput cost. Each is partial. Take the service offline when an unpatched or unsupported PHP serves an entry point to anonymous clients and no frontend rule can stop multipart or login-less access. CISA says to do this for systems that cannot be updated and for PHP 8.1 and earlier. If the service must stay up, put a login or VPN in front of every PHP path first: that removes the unauthenticated condition, and so makes the host a smaller target while a backport or a migration is prepared. A WAF rule is partial. Fastly's Next-Gen WAF has a CVE-specific rule for both CVEs, and Fastly warns that circulating IP block lists include VPN exits, Tor exits and residential addresses carrying ordinary traffic. ## CVE-2026-109618 and CVE-2026-109619 worked through, hour by hour PHP2Shell is two bugs in one release. CVE-2026-109618 (CVSS 9.8) corrupts request state on an out-of-memory failure, and CVE-2026-109619 (CVSS 8.1) leaves stale state that a later request on the same worker reads. CISA lists both in KEV. Fastly measured exploitation tools in its telemetry before the fix was public, so this case shows what the hour-by-hour plan can and cannot buy. All times are UTC. | When | What happened | Source | | --- | --- | --- | | 20 Oct 2026, about 15:00 | The CVE IDs circulate on forums and social media with a wrong release time and a wrong CVSS rating. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 20 Oct 2026, 19:30 | PHP publishes a pre-announcement and moves the coordinated release forward a day to 21 Oct at 14:00. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 21 Oct 2026, 03:59 | The first request from a PHP2Shell-named tool appears in Fastly's telemetry. The most widespread exploit kit follows at 05:11. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 21 Oct 2026, 07:41 | Chainguard publishes images containing the fix, 6 hours 19 minutes before the 14:00 release. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 21 Oct 2026, 13:34 | GitHub publishes the PHP advisories for both CVEs. | [source](https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w) | | 21 Oct 2026, 14:00 | PHP releases 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Fastly had seen more than 100 source IPs using PHP2Shell-named tools by then. NVD publishes the record at 14:17. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 21 Oct 2026 | CISA publishes an alert urging immediate patching of both CVEs, and names mod_php as the highest-risk server API. | [source](https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching) | | 22 Oct 2026, midday | Fastly has stored about 250,000 requests from about 3,200 public IPs, roughly a third labeled as defensive checks. | [source](https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell) | | 23 Oct 2026 | CISA updates the alert with a SIGMA detection rule and reinforced mitigation guidance. | [source](https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching) | | 26 Oct 2026 | CISA adds both CVEs to KEV with a due date of 29 Oct 2026. | [source](https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-109618) | Tools were scanning 9 hours and 35 minutes before the 13:34 advisory, so no plan beats the first requests, and Fastly says its counts cannot show which succeeded. What a plan buys is time relative to the other deadlines. Steps 1 to 4 need no fix. The pre-announcement at 19:30 on 20 Oct left 18 hours 30 minutes before the release for the SAPI inventory and the exposure order. From the advisory at 13:34, the 24-hour mark falls at 13:34 on 22 Oct, close to when Fastly's 250,000-request count was taken, and the 72-hour mark falls on 24 Oct, two days before the KEV entry and five days before its due date. A team that read only the CLI version, or restarted nothing after the package upgrade, would have carried the vulnerable worker through all of it. ## Related pages - [The communications playbook for the first 72 hours](/playbooks/communications/) - [The developer collaboration platform playbook for the first 72 hours](/playbooks/developer-collaboration-platform/) - [The executive playbook for the first 72 hours](/playbooks/executive/) ## Sources - PHP security advisory for CVE-2026-109618: https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 19 Nov 2026) - PHP security advisory for CVE-2026-109619: https://github.com/php/php-src/security/advisories/GHSA-6wmf-4rv5-p3rp (read 19 Nov 2026) - CISA alert on PHP vulnerabilities: https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching (read 19 Nov 2026) - CISA KEV entry for CVE-2026-109618: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-109618 (read 19 Nov 2026) - CISA KEV entry for CVE-2026-109619: https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-109619 (read 19 Nov 2026) - Fastly report on PHP2Shell exploitation and timeline: https://www.fastly.com/blog/how-fastly-detected-and-built-a-coordinated-response-to-php2shell (read 19 Nov 2026) - NVD record for CVE-2026-109618: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-109618 (read 19 Nov 2026) - NVD record for CVE-2026-109619: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-109619 (read 19 Nov 2026) Last reviewed 19 Nov 2026. Canonical: https://zerodayplaybook.com/playbooks/application-runtime/