F5 Rates Six BIG-IP iRule and APM Flaws Critical (CVE-2026-123637): Patch Now

F5 BIG-IP security advisory: six flaws rated Critical by F5 (CVSS 3.1 9.8), led by CVE-2026-123637. Patch now.

Status: F5 disclosed six unauthenticated flaws in BIG-IP on 18 November 2026; the CVE records give 10:00 EST as the public disclosure time. All six are stack-based buffer overflows (CWE-121) in the data plane. F5 rates each 9.8 (Critical) under CVSS 3.1 and 9.3 (Critical) under CVSS 4.0, says each lets a remote attacker without credentials make the Traffic Management Microkernel (TMM) terminate, and says each "may" allow remote code execution. None is reachable unless a specific feature is configured on a virtual server: one APM per-request policy combination (CVE-2026-123637) or one of five iRule commands. The only fixes are engineering hotfixes for BIG-IP 21.1.0, 17.5.x and 17.1.x, and engineering hotfix builds for BIG-IP Next for Kubernetes and BIG-IP CNF where those are affected. F5 lists no scheduled release with the fix and no mitigation, and warns that engineering hotfixes skip the full QA of scheduled releases. None of the six is in CISA's Known Exploited Vulnerabilities (KEV) catalog, F5's advisories do not report exploitation, and we found no public exploit code. On 19 November F5 revised five of the six advisories; the main change is that BIG-IP CNF is now listed as affected by five of the six flaws. We last checked at 02:20 EST on 19 November 2026.

Executive Summary

BIG-IP sits in front of the services that matter most: customer portals, remote access, APIs and DNS. Its Traffic Management Microkernel (TMM) is the process that carries that traffic. F5's six advisories describe the same failure in different features: crafted requests, which F5 does not describe, overflow a buffer on TMM's stack, TMM terminates, and traffic stops until it restarts. F5 says each flaw "may" also allow code execution, and the CVSS 3.1 vectors F5 assigned rate confidentiality, integrity and availability impact all high, over the network, with no privileges and no user interaction required.

Three things decide whether a BIG-IP is exposed:

  1. An affected build. BIG-IP 21.1.0, 17.5.0 to 17.5.1 or 17.1.0 to 17.1.3 without the 18 November engineering hotfix. For five of the six flaws (all except CVE-2026-123637), BIG-IP Next for Kubernetes and BIG-IP CNF 2.2.0 to 2.4.2 are also affected. F5 added CNF to three of those advisories on 19 November (see Technical Information). F5 does not evaluate releases past End of Technical Support (EoTS).
  2. The trigger feature on a virtual server. Each flaw needs a named configuration: an APM per-request policy that includes both the Client Information and Endpoint Security agents, or an iRule that uses PEM::subscriber, MR::message route, PSC::ip_address, GENERICMESSAGE::route or DNS::scrape, two of them only with an HTTP profile. A BIG-IP running none of these is not exposed to these six CVEs, whatever its version.
  3. Reachability of that virtual server. The attacker needs to send traffic to the virtual server that carries the feature. F5 says there is no control-plane exposure, so locking down the management interface does not help here; restricting who can reach the virtual server does.

How serious is it in practice? (Rescana guidance) These are narrower than a typical "every box is exposed" BIG-IP bug, because each needs a specific feature that an estate may not use at all. But where the feature is in use, the virtual server is by definition passing traffic, possibly from the internet, and F5 gives no workaround. F5's support articles say the BIG-IP system "temporarily fails to process traffic as it recovers from TMM restarting" and that high-availability pairs may fail over (K30911031, K15569543). The realistic minimum impact is therefore an outage of what the device fronts, not just the virtual server with the trigger, until TMM is back or a standby unit takes over. Treat any match as patch-now, and treat an unmatched estate as a configuration audit, not a patch emergency.

Exploitation status. CISA's KEV catalog (version 2026.11.18, released 14:17 EST, 1,786 entries) lists none of the six. CISA's Vulnrichment entries on all six CVE records rate exploitation "none", automatable "yes" and technical impact "total". FIRST's EPSS service had no score for any of them, and the NVD records were at status "Received", carrying only F5's scores. F5's advisories do not describe the triggering requests, and none of the sources we checked reports public exploit code or exploitation in the wild. CISA's "automatable: yes" is still worth noting: once a trigger is known, attacks against a reachable virtual server need no human in the loop.

Technical Information

The common pattern. Each advisory follows the same template: when a feature is configured on a virtual server, "undisclosed requests can cause the Traffic Management Microkernel (TMM) to terminate and may lead to remote code execution (RCE)". F5 says "Traffic is disrupted while the TMM process restarts" and "There is no control plane exposure; this is a data plane issue only." All six are classified CWE-121, stack-based buffer overflow. F5 has not published root-cause detail, and we do not speculate on it.

The six flaws and their triggers (per F5's advisories):

CVE F5 article Trigger on a virtual server Products F5 lists as affected Reported by
CVE-2026-123637 K000163937 APM provisioned, and a per-request policy that includes the Client Information agent and Endpoint Security agents BIG-IP APM Sina Kheirkhah, watchTowr
CVE-2026-121805 K000163946 An iRule that processes client traffic using PEM::subscriber BIG-IP PEM, BIG-IP CNF, BIG-IP Next for Kubernetes F5 (internal)
CVE-2026-120292 K000163940 An HTTP profile and an iRule using MR::message route BIG-IP (all modules), BIG-IP CNF, BIG-IP Next for Kubernetes Sina Kheirkhah, watchTowr
CVE-2026-120285 K000163939 An HTTP profile and an iRule containing PSC::ip_address BIG-IP (all modules), BIG-IP CNF, BIG-IP Next for Kubernetes Sina Kheirkhah, watchTowr
CVE-2026-116953 K000163941 An iRule that processes client traffic using GENERICMESSAGE::route BIG-IP (all modules), BIG-IP CNF, BIG-IP Next for Kubernetes Sina Kheirkhah, watchTowr
CVE-2026-120722 K000163945 An iRule that processes client traffic using DNS::scrape BIG-IP DNS, BIG-IP CNF, BIG-IP Next for Kubernetes F5 (internal)

The acknowledgements come from each F5 article. The Hacker News reports the same split: four from watchTowr's Sina Kheirkhah and two found by F5.

What the trigger features do. F5's iRules reference explains why these commands show up in some estates and not others:

  • PEM::subscriber creates, deletes or reads Policy Enforcement Manager subscriber sessions and needs a PEM licence. Expect it mainly at mobile and fixed-line service providers (Rescana context).
  • MR::message reads or sets entries in BIG-IP's message-routing table, and GENERICMESSAGE::route adds, deletes or looks up routes for the generic message protocol. Both belong to BIG-IP's Message Routing Framework, which F5's documentation says supports three protocol implementations: Diameter, SIP and Generic Message.
  • PSC::ip_address gets, sets or removes a list of PSC IP addresses.
  • DNS::scrape walks a DNS message and extracts fields from it. F5's reference says it needs the DNS profile, which comes only with BIG-IP DNS (GTM) or the DNS Services add-on.
  • APM Client Information and Endpoint Security agents inspect the connecting client, for example its device posture, inside a per-request policy. F5's advisory requires both in the same per-request policy for CVE-2026-123637.

How to read this (Rescana guidance): the five iRule flaws are most likely to matter in service-provider, signalling and DNS deployments, while CVE-2026-123637 concerns enterprises that use APM for remote access or zero-trust posture checks. In both groups the trigger sits on a production virtual server, so the exposure is real where it exists.

F5's 19 November revision: BIG-IP CNF. When F5 published the advisories on 18 November, K000163940 (CVE-2026-120292), K000163941 (CVE-2026-116953) and K000163945 (CVE-2026-120722) listed BIG-IP CNF as "Not vulnerable". F5 updated all three on 19 November: each now lists BIG-IP CNF 2.4.0 to 2.4.2, 2.3.0 to 2.3.4 and 2.2.0 to 2.2.5 as affected, rated Critical (9.8 under CVSS 3.1, 9.3 under CVSS 4.0), with fixes in 2.4.2-EHF1, 2.3.4-EHF1 and 2.2.5-EHF1, and K000163945 now names BIG-IP CNF in F5's tracking ID 2535409. That removes the earlier mismatch for CVE-2026-120722, whose CVE record already listed "BIG-IP Next CNF" as affected. The CVE records for CVE-2026-120292 and CVE-2026-116953 still did not list CNF when we checked at 02:16 EST on 19 November; we follow the newer F5 articles. CNF operators whose iRules use MR::message route, GENERICMESSAGE::route or DNS::scrape are in scope and should plan the CNF hotfix (Rescana guidance).

Related but different: CVE-2026-94127. On 22 September 2026, CISA added CVE-2026-94127 to the KEV catalog. CISA describes it as a heap-based buffer overflow (CWE-122) in BIG-IP APM when an access policy and an OAuth profile are configured on a virtual server. Today's six are separate stack-based flaws with different triggers. It is a reminder that configuration-specific BIG-IP APM bugs have been exploited before, not evidence about these six.

Fixed alongside them. F5's out-of-band security notification K000164058 indexes 22 advisories published the same day, 10 of them for NGINX, including the HTTP/2 flaw we covered in our NGINX CVE-2026-123007 advisory. F5 rates six more BIG-IP and BIG-IQ issues High under CVSS 3.1. Three APM TMM crashes are fixed in the same engineering hotfixes: CVE-2026-125272 (form-based Logon Page in a per-request policy), CVE-2026-126822 (SAML Service Provider access profile) and CVE-2026-121812 (the tmm.access.policytrace key, off by default; F5 says it needs an authenticated attacker, and it also affects SSL Orchestrator). Three more are fixed in scheduled releases 21.1.0.3, 17.5.1.10 and 17.1.3.6: CVE-2026-126336 (APM SAML SP agent with encrypted assertions), CVE-2026-125274 (authenticated iControl SOAP, 8.8 under CVSS 3.1, also BIG-IQ, fixed in 8.4.2.3) and CVE-2026-120718 (iControl REST tokens that may not expire as expected; F5 scores it 8.1 (High) CVSS 3.1 / 9.2 CVSS 4.0, and its 21.x fix is 21.1.0.2).

Plan the batch as one change (Rescana guidance): the 18 November engineering hotfixes cover the six Critical-rated flaws and three of the High APM issues directly, and they are built on base releases at or above the scheduled BIG-IP fixes for the rest. BIG-IQ needs its own update to 8.4.2.3 for CVE-2026-125274. Confirm that in F5's release notes for your build.

Affected Product Versions

Product and branch Affected (per F5's six advisories) Fixed in (F5's tables)
BIG-IP 21.x 21.1.0 BIGIP-21.1.0.4.0.19.3-ENG
BIG-IP 17.5.x 17.5.0 to 17.5.1 BIGIP-17.5.1.11.0.61.12-ENG or BIGIP-17.5.1.10.0.80.35-ENG
BIG-IP 17.1.x 17.1.0 to 17.1.3 BIGIP-17.1.3.7.0.21.4-ENG or BIGIP-17.1.3.6.0.14.20-ENG
BIG-IP Next for Kubernetes 2.4.x / 2.3.x / 2.2.x (all except CVE-2026-123637) 2.4.0 to 2.4.2, 2.3.0 to 2.3.4, 2.2.0 to 2.2.5 2.4.2-EHF1, 2.3.4-EHF1, 2.2.5-EHF1
BIG-IP CNF 2.4.x / 2.3.x / 2.2.x (all except CVE-2026-123637) 2.4.0 to 2.4.2, 2.3.0 to 2.3.4, 2.2.0 to 2.2.5 2.4.2-EHF1, 2.3.4-EHF1, 2.2.5-EHF1

Notes on the table:

  • Module scope. CVE-2026-123637 affects BIG-IP APM only, CVE-2026-121805 BIG-IP PEM only and CVE-2026-120722 BIG-IP DNS only. The other three are listed for "BIG-IP (all modules)", because any module can run an iRule. F5 lists BIG-IP Next for Kubernetes and BIG-IP CNF as not vulnerable to CVE-2026-123637.
  • Which 17.x build. F5's article K000164239 maps each build to what you run now: BIGIP-17.1.3.6.0.14.20-ENG only if you already run BIGIP-17.1.3.6.0.9.20-ENG; BIGIP-17.5.1.10.0.80.35-ENG only if you run BIGIP-17.5.1.10.0.7.35-ENG or 17.5.1.10.0.37.35-ENG; otherwise 17.1.3.7.0.21.4-ENG, 17.5.1.11.0.61.12-ENG or 21.1.0.4.0.19.3-ENG for systems on the 17.1.3.7, 17.5.1.11 or 21.1.0.4 base release or earlier. A system on any other custom engineering hotfix needs a new one from F5 Support.
  • QA caveat. F5's footnote on every fix says engineering hotfixes "do not undergo the extensive QA assessment of scheduled software releases" and that F5 offers them "with no warranty or guarantee of usability". There is no scheduled release with these fixes yet.
  • Kubernetes and CNF build names. Since 19 November, F5's articles give the same build names for BIG-IP Next for Kubernetes and BIG-IP CNF (for example 2.4.2-EHF1) and say these hotfixes come from the F5 Artifact Repository. F5's CVE records still use the earlier BNK-prefixed names, such as BNK-2.4.2-EHF1, for both products.
  • End of Technical Support. F5 evaluates only versions that have not reached EoTS. A release absent from the tables, such as 16.1.x or 15.1.x, has not been shown to be safe; treat it as exposed where the trigger feature is configured, and plan a move to a supported branch (Rescana guidance).
  • Not affected. F5 lists BIG-IQ, F5OS, F5OS-A, F5OS-C, F5 Distributed Cloud, NGINX One Console, NGINX and F5 AI Gateway as not vulnerable to all six.

How to Find Your Exposure

The steps below are Rescana guidance unless a source is named. All are read-only.

1. Inventory versions and modules. On each BIG-IP, tmsh show sys version gives the release and any engineering hotfix, and tmsh list sys provision one-line shows which modules are provisioned (look for apm, pem and gtm (BIG-IP DNS) with a level other than none). F5's advisories also point to its iHealth service for diagnosing vulnerabilities on BIG-IP systems.

2. Search every iRule for the five commands. iRules live in partitions other than /Common, so search recursively from the root:

tmsh -q -c 'cd /; list ltm rule recursive' > /var/tmp/irules.txt grep -nE 'PEM::subscriber|MR::message[[:space:]]+route|PSC::ip_address|GENERICMESSAGE::route|DNS::scrape' /var/tmp/irules.txt

A match shows the rule; you still need to know whether it is attached. Also search the saved configuration directly, which catches rules created by iApps or automation: grep -lE 'PEM::subscriber|MR::message|PSC::ip_address|GENERICMESSAGE::route|DNS::scrape' /config/bigip.conf /config/partitions/*/bigip.conf.

3. Map rules to virtual servers. tmsh -q -c 'cd /; list ltm virtual recursive rules profiles destination' lists each virtual server's iRules, profiles and listening address. For CVE-2026-120292 and CVE-2026-120285, F5's trigger also needs an HTTP profile on the same virtual server. Note which matching virtual servers listen on internet-facing or partner-facing addresses.

4. Check APM per-request policies. In the Configuration utility, open each per-request policy under Access in the Visual Policy Editor and look for both a Client Information item and an Endpoint Security item; F5's trigger needs both in the same policy. Then confirm which virtual servers use that policy: in tmsh, a virtual server's per-request policy is its per-flow-request-access-policy property (F5 tmsh reference), so tmsh -q -c 'cd /; list ltm virtual recursive per-flow-request-access-policy' shows the attachments.

5. Kubernetes and CNF. For BIG-IP Next for Kubernetes and BIG-IP CNF, export the iRule resources your deployment applies (from Git or with kubectl get ... -o yaml for your iRule custom resources) and run the same search. Record the BNK or CNF version for each cluster.

6. Prioritise. Patch first where a matching virtual server is reachable from the internet or from partner networks, then devices that front many services, because a TMM crash takes all of them down at once. Devices with no matching configuration can take the hotfix in a normal window, mainly for the High-rated issues in the same batch.

7. Confirm after upgrading. tmsh show sys version should show the hotfix build. On a device pair, upgrade the standby first, fail over, verify traffic, then upgrade the former active unit.

Workaround and Mitigation

Vendor position. All six F5 articles list the mitigation as "None". The fix is the engineering hotfix.

Interim options (Rescana guidance, not F5-published mitigations). Because F5 ties each flaw to a feature, removing the feature removes the precondition F5 names:

  • iRules. Where a matching iRule is not essential, detach it from the virtual server or rewrite it without the command. Test first: for PEM, message routing or DNS, the command is usually doing real work, and removing it changes behaviour. Keep the change in configuration management so it can be reversed after the hotfix.
  • APM per-request policies. Where both agents are present, consider whether the posture check can move out of the per-request policy until you patch. That weakens a security control, so it is a risk decision, not a default.
  • Limit who can reach the virtual server. Where a matching virtual server serves a known set of clients, such as SIP or Diameter peers, restrict its source addresses with an upstream ACL or BIG-IP packet filter. It does not help for internet-wide services.
  • Plan for a crash. Make sure device pairs fail over cleanly on a TMM restart. That softens one crash, not repeats: the peer runs the same vulnerable configuration. F5 recommends high availability to lessen the impact of a related APM crash bug in the same batch, CVE-2026-126336.

Handling the reduced-QA risk. F5's own warning is the reason to stage, not to wait: install the hotfix on a lab or standby unit first, run your regression for the matching features, keep the prior boot location so you can roll back, and watch the device for TMM restarts and support cases after the change.

Indicators of Compromise

F5 has published no indicators of compromise, and no exploitation has been reported to derive them from. The items below are detection leads, not IoCs (Rescana guidance).

  • TMM restarts. A crash of TMM is the flaw's minimum visible effect. F5's support articles say a TMM crash writes a core file to /shared/core, with messages in /var/log/tmm and /var/log/ltm (K30911031, K16197). Watch those locations, and unexpected failovers on device pairs, especially on devices with a matching virtual server. Alert on any TMM restart until you have patched.
  • Correlate with traffic. If TMM restarts, note which virtual servers were receiving traffic at that moment and from where. Repeated restarts that line up with one client source against a matching virtual server deserve escalation to F5 Support.
  • Unexpected change. Code execution in TMM, if achieved, would sit in the process that holds traffic and TLS keys. Review configuration change logs, /var/log/audit and new or altered iRules after any unexplained TMM crash.
  • If you suspect compromise. Collect a QKView and core files for F5 Support before rebooting, then rotate TLS keys and certificates served by the device and credentials stored in its configuration.

Third-Party Risk (TPRM) Angle

Suppliers may run BIG-IP in front of the services you depend on: SaaS login and API front doors, managed remote-access portals, service-provider signalling, and DNS. You cannot see their iRules or APM policies, and you cannot patch their devices. What you can do is find out which suppliers have a matching configuration and hold them to a date (Rescana guidance).

The risk concentrates in three supplier types:

  • SaaS and hosting providers and MSPs that run BIG-IP LTM or APM in front of customer services. A TMM crash takes down every service on that device; code execution would put an attacker in the traffic path.
  • Telecoms, mobile and signalling providers that run PEM, message routing or BIG-IP Next for Kubernetes and CNF, where most of the iRule triggers are likely to appear.
  • DNS providers and managed DNS services on BIG-IP DNS that use DNS::scrape, including those on BIG-IP CNF, which F5 has listed as affected since 19 November.

Send these three questions to the suppliers and MSPs that run BIG-IP in front of customer services (Rescana guidance):

  1. Does any virtual server in your estate use an APM per-request policy containing both the Client Information and Endpoint Security agents, or an iRule using PEM::subscriber, MR::message route, PSC::ip_address, GENERICMESSAGE::route or DNS::scrape?
  2. On what date will you install F5's 18 November engineering hotfix on affected devices, and how are you handling F5's warning that engineering hotfixes skip the full QA of scheduled releases (staging, regression testing, rollback plan)?
  3. Until then, how are you monitoring TMM availability, and is each affected device in a high-availability pair that fails over cleanly if TMM restarts?

Ask for evidence, not reassurance: the BIG-IP version and hotfix build on the devices that carry your traffic, confirmation that the search in steps 2 to 4 above was run across all partitions, and a commitment to tell you within an agreed window about unexplained TMM crashes. Record answers against each vendor record, and set follow-ups for suppliers on EoTS releases, on custom engineering hotfixes waiting for F5 Support, and on BIG-IP CNF, which F5 added to three of the advisories on 19 November.

What to do internally: search your own BIG-IP estate for the six triggers, hotfix matching internet-facing devices first, stage the hotfix to manage the QA risk, and alert on TMM restarts. We will update this advisory if CISA adds any of the six to KEV, exploitation or public exploit code is reported, F5 ships a scheduled release with the fixes, or F5 changes the affected products or fix builds again.

Need help working out which suppliers run BIG-IP configurations exposed to CVE-2026-123637 and the five iRule flaws? Our team can help you prioritize them and track their answers. Talk to an expert

Sources

Talk to an expert

Talk to Rescana about this advisory — or about how we help TPRM teams keep reviews moving.