CVE-2026-94127: unauthenticated RCE in F5 BIG-IP APM's OAuth authorization server
A heap overflow in BIG-IP APM (CVSS 9.8, CISA KEV) gives unauthenticated RCE when APM acts as an OAuth authorization server. Who is exposed, F5's IOCs, hotfixes and the iRule stopgap.
F5 disclosed CVE-2026-94127 on 22 September 2026 as an already-exploited zero-day, and CISA added it to the KEV catalog the same day. It’s a heap-based buffer overflow in BIG-IP Access Policy Manager (APM) that gives an unauthenticated attacker remote code execution. The good news is that it needs a specific configuration. The bad news is that the configuration is a common one for organizations using APM as their single sign-on front door. This is the first of three CISA KEV entries I’m covering this week.
Scope and ethics. This post covers architecture, exposure checks, detection and remediation for systems you own or are authorized to assess. It contains no trigger traffic or exploit code. No public proof of concept had been confirmed at the time of Rapid7’s write-up, and nothing here changes that.
At a glance
| CVE | CVE-2026-94127 |
| Severity | CVSS 3.1 9.8 Critical AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H; CVSS 4.0 9.3 (F5 SIRT, via NVD) |
| CWE | CWE-122 Heap-based Buffer Overflow |
| Affected | BIG-IP APM 21.1.0, 17.5.0–17.5.1, 17.1.0–17.1.3, only when APM is an OAuth authorization server |
| Fixed in | Engineering hotfixes: Hotfix-BIGIP-21.1.0.2.0.30.22-ENG, Hotfix-BIGIP-17.5.1.9.0.160.12-ENG, Hotfix-BIGIP-17.1.3.5.0.41.14-ENG |
| Exploited? | Yes. F5: “this vulnerability has been exploited.” CISA KEV, added 22 Sep 2026 (federal due date 25 Sep), forensic triage required |
| Auth / interaction | None / none |
Description: how the bug works
BIG-IP APM is F5’s access gateway. It sits in front of applications on a BIG-IP virtual server, runs an access policy (login, MFA, posture checks) and can act in several OAuth roles: as a client or resource server that consumes tokens, or as an authorization server that issues and validates them for other applications.
The flaw lives in the traffic-processing path of that last role. F5’s description: “When a BIG-IP APM access policy and an OAuth profile are configured on a virtual server, specific malicious traffic can lead to remote code execution.” F5 narrowed the CVE record on 23 September to say the bug “is only present when BIG-IP APM is configured as an OAuth Authorization Server,” and that deployments using APM “strictly as an OAuth Client / Resource Server” are not affected.
F5 hasn’t published the exact parsing routine. What is public is the class: a heap-based buffer overflow (CWE-122). In general terms, that means request data gets written past the end of a heap buffer, corrupting neighboring memory in the process handling the request. My reading of F5’s indicators (below) is that the UserInfo request path and TMM, the Traffic Management Microkernel that handles data-plane traffic, are involved, but F5 hasn’t confirmed which function overflows. Because the bug is in the data plane, attackers reach it through the same virtual server your users log in through, not the management interface.
Impact
- Unauthenticated remote code execution on the BIG-IP. An access gateway sees every user’s credentials and session, holds TLS private keys, and usually has routes into internal networks. Treat a compromised one as a compromised identity tier.
- Appliance mode doesn’t help. F5 states systems in Appliance mode are also vulnerable.
- Locking down the management port doesn’t help either. F5 calls this “a data plane issue; there is no control plane exposure,” which means the attack path is the public-facing virtual server.
Which deployments are exposed
| Configuration | Status |
|---|---|
| APM access policy + OAuth authorization server profile on the same virtual server, affected branch | Vulnerable |
| APM as OAuth client or resource server only | Not affected (per F5) |
| APM without any OAuth profile | Not affected (needs both pieces) |
| BIG-IP 17.1.3 / 17.5.1.3 (the CVE-2025-53521 fix builds) | Still need this hotfix if APM is an authorization server (The Hacker News, citing F5) |
| End of Technical Support branches | Not evaluated by F5. Assume vulnerable if the configuration matches |
Prerequisites: network reachability to the APM virtual server, an affected version, and the authorization-server configuration. No credentials, no user interaction. Shadowserver reported over 14,700 IPs with BIG-IP APM fingerprints at disclosure (via BleepingComputer); that’s all APM, not only vulnerable configurations.
Identifying affected systems
On each BIG-IP you administer, check the version and installed hotfixes, then look for authorization-server profiles and the virtual servers they’re attached to. Command syntax can vary slightly across TMOS releases, so check it against your version’s tmsh reference.
# 1) Version and installed hotfixes (look for the -ENG hotfix names in the table above)
tmsh show sys version
tmsh show sys software status
# 2) Is APM provisioned at all?
tmsh list sys provision apm level
# 3) OAuth authorization-server profiles (empty output = you're not an authorization server)
tmsh list apm oauth oauth-profile
# 4) Which virtual servers carry an access profile? Cross-reference with (3)
tmsh list ltm virtual one-line | grep -iE 'access|oauth'
If step 3 returns profiles and step 4 shows them on a virtual server with an access policy, and the hotfix isn’t installed, you’re exposed. Use the F5 GUI (Access > Federation > OAuth Authorization Server) to double-check if the tmsh output is ambiguous.
Rapid7 Exposure Command, InsightVM and Nexpose have authenticated checks for this CVE (added in the 23 September content release).
Safe validation in an isolated lab
You don’t need to send malicious OAuth traffic to prove you’re fixed. A useful lab exercise checks that your inventory, detection and patch process works:
- On a BIG-IP VE in an isolated lab, build an APM access policy with an OAuth authorization-server profile on an affected build.
- Run the inventory commands above and your vulnerability scanner. Both should flag it. Then apply the hotfix and confirm both stop flagging it.
- Generate benign failed UserInfo requests with an invalid token (a normal client misconfiguration) and confirm they appear in
/var/log/apmand that your SIEM rule for the threshold below fires. - Confirm your baseline for
tmctl global_oauth_statso you know what “normal”total_failedlooks like.
Detection and threat hunting
F5’s indicators (relayed in CERT-EU’s advisory and reported by The Hacker News) are behavioral. There are no hashes or IPs. No single indicator proves compromise; the sequence does.
- APM log,
/var/log/apm. Repeated failed UserInfo requests with the error “The access token is invalid.” Ten or more from one source IP in a short period is the pattern to flag.# count invalid-token failures per source; review anything spiky grep -h "access token is invalid" /var/log/apm* | grep -oE '([0-9]{1,3}\.){3}[0-9]{1,3}' | sort | uniq -c | sort -rn | head - OAuth counters. An unexplained rise in
total_failed:tmctl global_oauth_stat -s total_requests,total_userinfo_requests,total_failed - Audit log,
/var/log/audit. Suspicious commands at around the same time as the authentication failures. - TMM aborts. F5 has seen TMM enter a loop that causes the SOD (failover) daemon to send it a SIGABRT. Check
/var/log/ltmfor TMM restarts and/var/corefor new TMM core files. A core file alone isn’t evidence. - The combination of repeated OAuth failures, then suspicious commands, then a TMM SIGABRT shortly afterwards warrants human review and, in my view, an incident response.
Also hunt the usual post-exploitation signs on any appliance: new local accounts, changed iRules or access policies, unexpected cron jobs or processes, and outbound connections from the BIG-IP’s self IPs.
Preserve evidence before you patch. CISA’s KEV entry requires forensic triage, and CERT-EU recommends collecting evidence (a qkview, logs and any core files) before installing the hotfix. F5, CISA and CERT-EU haven’t said whether the hotfix removes access an attacker already gained, so I wouldn’t assume it does.
Remediation
- Install the engineering hotfix for your branch:
Hotfix-BIGIP-21.1.0.2.0.30.22-ENG,Hotfix-BIGIP-17.5.1.9.0.160.12-ENGorHotfix-BIGIP-17.1.3.5.0.41.14-ENG. Confirm the names against K000162605 before you download. - Collect evidence first if the system was exposed before 22 September, and hunt using the indicators above.
- EoTS branches: upgrade to a supported branch. F5 didn’t evaluate them, so there’s no assurance either way.
- After patching an exposed unit, rotate what the box holds: TLS keys and certificates, OAuth signing keys and client secrets, local and admin passwords, and any directory bind credentials.
Interim mitigation (if you can’t patch today)
- F5’s iRule. F5 provides an iRule for the affected virtual server, available only by opening a case with F5 Support (BleepingComputer cites K000135931). CISA’s KEV note tells agencies to apply it first “to allow for proactive forensic triage,” then install the final patch.
- If the authorization-server role isn’t business-critical, remove the OAuth authorization-server profile from internet-facing virtual servers until patched. Test this in a maintenance window, since it will break OAuth sign-in for dependent apps.
- Restricting the management interface does not mitigate this bug.
References
- F5 advisory K000162605 (login may be required)
- CVE-2026-94127, NVD · CISA KEV catalog entry
- Rapid7 emergent threat response
- The Hacker News: versions, F5/CERT-EU indicators
- BleepingComputer: exploitation, iRule, Shadowserver exposure
- Field Effect analysis
Verify every version against the vendor advisory before acting. Published 04 Nov 2026.