Skip to main content
CybersecurityVulnerability Management

F5 BIG-IP Flaws Expose Networks to Crash, Code Execution Risks

Network operations room with central router or switch surrounded by cables and monitoring equipment.

“They may also allow remote code execution,” F5 warned today after disclosing six critical vulnerabilities in BIG‑IP that can crash the Traffic Management Microkernel (TMM) without logging in and, in specific configurations, may permit remote code execution.

The six CVEs and the immediate technical risk

F5 rates all six flaws 9.8 under CVSS v3.1 and 9.3 under CVSS v4.0. According to the company's advisories, each is a stack‑based buffer overflow in BIG‑IP’s data plane — the component that processes almost all network traffic — and each can let an unauthenticated attacker crash TMM on an affected virtual server. F5 also says the management side is not exposed. None of the advisories reports known exploitation.

Exact configurations that trigger each vulnerability

  • CVE‑2026‑123637 — requires an APM per‑request policy that includes both the Client Information agent and the Endpoint Security agent. Affects BIG‑IP APM only.
  • CVE‑2026‑121805 — requires an iRule that uses PEM::subscriber on client traffic. Affects BIG‑IP PEM, BIG‑IP Next for Kubernetes, and BIG‑IP CNF (CNF marked affected as of today).
  • CVE‑2026‑120292 — requires an HTTP profile and an iRule that uses MR::message route. Affects BIG‑IP (all modules) and BIG‑IP Next for Kubernetes.
  • CVE‑2026‑120285 — requires an HTTP profile and an iRule that uses PSC::ip_address. Affects BIG‑IP (all modules), BIG‑IP Next for Kubernetes, and BIG‑IP CNF (CNF marked affected as of today).
  • CVE‑2026‑116953 — requires an iRule that uses GENERICMESSAGE::route on client traffic. Affects BIG‑IP (all modules) and BIG‑IP Next for Kubernetes.
  • CVE‑2026‑120722 — requires an iRule that uses DNS::scrape on client traffic. Affects BIG‑IP DNS and BIG‑IP Next for Kubernetes. F5’s advisory lists BIG‑IP CNF as not vulnerable, but the CVE record updated today lists CNF as affected.

Patching: engineering hotfixes only, with caveats

F5 has released engineering hotfixes as the only fixes for the six critical flaws; the advisories list no scheduled release as a fix and offer no workaround until a hotfix is installed. The company warns that “engineering hotfixes do not undergo the extensive QA assessment of scheduled software releases,” and its hotfix policy states they come with no warranty or promise that they will work. Systems already patched with a custom engineering hotfix will need a new hotfix issued by F5 Support.

For BIG‑IP Next for Kubernetes, the fixes are provided in BNK‑2.4.2‑EHF1, BNK‑2.3.4‑EHF1 and BNK‑2.2.5‑EHF1 for the respective 2.4.0–2.4.2, 2.3.0–2.3.4 and 2.2.0–2.2.5 version ranges. BIG‑IP CNF is fixed for CVE‑2026‑121805 and CVE‑2026‑120285 in CNF‑2.4.2‑EHF1, CNF‑2.3.4‑EHF1 and CNF‑2.2.5‑EHF1 on the same version ranges. F5 checks only versions that remain within technical support, so older releases not listed have not been shown to be safe.

Six other BIG‑IP and BIG‑IQ fixes published today

Alongside the six critical data‑plane flaws, F5 published advisories for six lower‑rated vulnerabilities in BIG‑IP and BIG‑IQ. Highlights from the company’s CVSS v3.1 scores and mitigation notes include:

  • CVE‑2026‑125274 (iControl SOAP, 8.8) — a user with network access and at least the Guest role can send an unauthorized management request; fixed in BIG‑IP 21.1.0.3, 17.5.1.10 and 17.1.3.6, and BIG‑IQ 8.4.2.3. The SOAP advisory explains how to limit iControl SOAP access to trusted networks and warns that blocking it on self‑IP addresses may break high‑availability setups.
  • CVE‑2026‑120718 (iControl REST, 8.1) — iControl REST tokens may not expire as expected; fixed in BIG‑IP 21.1.0.2, 17.5.1.10 and 17.1.3.6. As a workaround, scripts that create tokens can be changed to delete them when no longer needed.
  • CVE‑2026‑125272, CVE‑2026‑126822 and CVE‑2026‑126336 (APM and SAML issues, 7.5) — these can crash TMM in specific APM access profile setups; they are fixed in the same engineering hotfixes as the six critical flaws or in listed scheduled releases. For one SAML issue, F5 suggests removing the SAML Auth agent if SAML SP is not needed and limiting access to trusted client networks; for another, F5 recommends a high‑availability setup to reduce impact.
  • CVE‑2026‑121812 (APM and SSL Orchestrator, 7.5) — a per‑request policy with the tmm.access.policytrace database key turned on can crash TMM; the key is off by default. APM is fixed in the same engineering hotfixes and the SSL Orchestrator advisory lists fixes on 21.1 and 17.1 branches.

What this means for technologists, procurement leaders, and adversaries

  • Technologists and security teams: prioritize installing F5’s engineering hotfixes for affected, in‑support versions and audit virtual servers for the specific iRule and APM per‑request configurations named in the advisories, since no workaround is provided for the critical flaws.
  • Procurement and operations leaders: note that F5’s hotfix policy warns about limited QA and no warranty; teams running custom hotfixes will need new builds from support, and older, unsupported product versions have not been verified safe.
  • Adversaries and threat analysts: while F5 reports no exploitation yet, the company contrasts these stack‑based overflows with CVE‑2026‑94127 — a heap‑based APM flaw disclosed in September that was later exploited and added to CISA’s Known Exploited Vulnerabilities catalog — underscoring the operational incentive to watch these CVEs and affected configurations closely.

F5 discovered two of the six critical flaws itself; a WatchTowr researcher, Sina Kheirkhah, reported the other four. The company’s broader security notification published today includes 22 advisories in total, ten of them for NGINX. For administrators, the immediate choice is stark: apply engineering hotfixes from F5 Support for affected, in‑support versions or continue to run configurations that the advisories identify as trigger points for TMM crashes and possible remote code execution.

Original story