Skip to main content

F5 Fixes Two NGINX HTTP/2 Flaws That Let an Unauthenticated Attacker Run Code, One Dating Back to 2015

F5's out-of-band NGINX release fixes two critical-rated HTTP/2 flaws, CVE-2026-123007 and CVE-2026-123633. Fixed versions, why a version check can mislead, and what turning HTTP/2 off does and does not cover.

Close-up of yellow fiber optic cables in a network server, showcasing fast data transfer.
KBY News Desk9 min read

In this article

Share

F5 shipped an out-of-band security release for open-source NGINX on Wednesday 18 November, fixing ten vulnerabilities, two of which let an unauthenticated attacker run code on a server that has HTTP/2 switched on. The fixed versions are 1.31.9 on the mainline branch and 1.30.7 on the stable branch, according to the project’s own advisory list. One of the two flaws, CVE-2026-123007, is in code that has been in every HTTP/2-capable release since 1.9.5 in 2015. F5 told The Register it has found no evidence that any of the flaws is being exploited, and none is in CISA’s Known Exploited Vulnerabilities catalogue as of its latest edition. Working proof-of-concept code is already public for at least one of them, so that position may not last.

#What was fixed, and by whom it was found

The nginx.org security advisories page lists ten new entries, all fixed in 1.31.9 and 1.30.7 (one, a null pointer dereference in PROXY protocol handling, only ever affected the 1.31 line and is fixed in 1.31.9). The two HTTP/2 flaws were each found independently by several groups: Wiz Research, Ikotas Labs, Metnir, Rapid7 and Xint all reported at least one of them to F5, and Wiz says F5 told it another team had already reported CVE-2026-123633 when it submitted its own report. Wiz and Xint both say AI tooling was part of how the bugs were found. Xint says one of its researchers found CVE-2026-123007 while preparing for the ZeroDay Cloud 2026 contest, and that a fully automated scan found four more a few days later.

The two that matter most are both memory-safety bugs in how NGINX handles HTTP/2 traffic. In CVE-2026-123633, Wiz says, a connection’s memory can be freed while a request still needs it, a use-after-free. In CVE-2026-123007, flow-control accounting goes wrong and NGINX ends up copying data into the wrong place. Wiz demonstrated code execution against the second on a server doing nothing more than serving static files over HTTPS on a 64-bit system, with no backend application and no third-party modules. It also says ordinary memory protections such as address space randomisation do not stop either bug, because an attacker can read memory before running code. We are not repeating the technique; the researchers’ write-ups explain it for defenders, and that is enough to act on.

#Which servers are in scope

CVE-2026-123007 affects every official release from 1.9.5 to 1.31.8 and needs HTTP/2 to be enabled. CVE-2026-123633 affects 1.19.7 to 1.31.8, needs HTTP/2, and applies when NGINX is acting as a reverse proxy and its connection slots run low, according to Wiz. A server built without the HTTP/2 module, or one with HTTP/2 off, is outside both. Xint’s survey of Shodan on 10 November counted 56.7 million hosts reporting NGINX, of which 11.94 million showed an identifiable version, and 98.9 per cent of those fell inside the affected range. Xint is careful to say that is a count of version strings, not of vulnerable servers: some will have HTTP/2 off, and some will carry a distribution’s backported fix under an old version number.

That last point is the one most likely to trip up a patching check. Xint notes that Ubuntu 24.04 ships NGINX 1.24.0, Ubuntu 22.04 ships 1.18.0 and Debian 12 ships 1.22.1, and that distributions often fix a flaw without changing the number that nginx -v prints. Its example is Ubuntu 22.04, where the package moved from 1.18.0-6ubuntu14 to 1.18.0-6ubuntu14.22 while NGINX still reported 1.18.0. For a distribution package, compare the full package version with the distributor’s security notice. For commercial NGINX Plus, Wiz’s table lists fixed releases of 37.0.7.1 and 37.2.0.1 for CVE-2026-123633 and 37.0.8.1 and 37.3.0.1 for CVE-2026-123007; check F5’s own advisory for your branch.

#Why the ratings do not agree

The project’s page rates CVE-2026-123007 “major” and CVE-2026-123633 “medium”. Wiz scores both 9.2 out of 10 under CVSS 4.0 and calls them critical, and Xint calls the full set of five critical. The Register reports them as critical, too. The difference matters because “medium” is what a patch queue might sort to the bottom. Where the project and the researchers disagree we follow the project for what is affected and the researchers for what has been demonstrated: Wiz and Xint have shown code execution, in Xint’s case on cloud instances and bare metal, and Rapid7 has published a proof of concept for CVE-2026-123007 against a Linux Docker image running 1.31.8, according to The Register. A bug with a public exploit that gives code execution before login is not a medium problem on an internet-facing server, whatever the label.

Xint also says it chained CVE-2026-123007 with a separate flaw in the Linux kernel’s SCTP code, present since 2018, to reach root on the host and escape a container. That chain is Xint’s, and the kernel bug is a different fix to track. In our reading, it shows what an attacker can hope for after the first step: code running as the NGINX worker is a foothold, and a container is not a boundary to rely on if the host kernel is unpatched.

#Turning HTTP/2 off helps, but only part of the way

If you cannot upgrade today, Wiz and Xint give the same temporary measure. On NGINX 1.25.1 or later, set http2 off; in each affected server block and remove any http2 parameter from listen lines. On 1.9.5 to 1.25.0, remove http2 from the listen line. Validate the configuration, reload, and confirm the old workers have exited. Clients fall back to HTTP/1.1, which most will not notice, but anything that needs HTTP/2 end to end, native gRPC for example, will break.

Two cautions from Xint. First, this blocks only CVE-2026-123007 and CVE-2026-123633. The other three critical-rated bugs it found are in other places: a resolver bug that needs a request-derived proxy_pass destination, an HTTP/3 proxying bug, and one in the mail proxy module. They are untouched by turning HTTP/2 off, though each needs a less common configuration. Second, installing the update is not finished by a reload: Xint says to restart NGINX or complete a supported live binary upgrade, because a reload does not replace the running executable. That is easy to miss and leaves a patched package on disk and a vulnerable process in memory.

#How to tell whether anyone has tried it

Do not expect the access log to help much. Xint says the HTTP/2 control frames used to trigger CVE-2026-123007 are not logged as requests and are encrypted in transit. It suggests four places to look instead. The clearest is any program started by an NGINX worker, which has no ordinary reason to launch a shell; on a host with auditd, an execve rule keyed to the worker’s user ID will record it, though only from the moment the rule is added. The others are weaker: a long-lived connection from one address that receives far more than it sends, many short TLS connections from the same address, and repeated worker crashes in the error log. Any of those can have an innocent cause, so look for several together.

Xint also provides a quick external test: openssl s_client with the -alpn h2 option shows whether a host offers HTTP/2. If a CDN sits in front, the test only reaches the CDN, so run it against the origin as well.

#Our view: patch exposed servers this week, then find the NGINX you did not know you ran

We think this deserves the same treatment as an exploited flaw on anything reachable from the internet with HTTP/2 on, even though nobody says it is exploited yet. Our reasoning is simple: the researchers have published enough for a competent attacker to build on, five teams found the same bugs independently, and the affected version range covers close to every server people have heard of. We expect exploitation attempts to appear within days of the proof of concept spreading; the evidence that would show us wrong is a quiet fortnight with no reports from F5, CISA or the honeypot operators. That is a prediction, not a fact.

The harder part is inventory. NGINX is rarely just “the web server”. It is the front door in container images, ingress controllers, load balancers and appliances that bundle their own copy, and a supplier’s version string may say nothing about the bundled build. The question to put to each supplier, in these words, is: “Which NGINX release does this product contain, does it include the HTTP/2 fixes released on 18 November, and when will you ship them?” A vendor that cannot answer within a day has told you something about how it tracks its own dependencies.

There is also a pattern worth naming. This follows the Java Base64 flaw we covered on Wednesday, the PHP bugs Wiz reported last month and a critical Apache flaw The Register says is under active exploitation. Wiz’s own framing is that AI makes old memory-safety bugs cheaper to find for everyone, defenders and attackers alike. We would plan for patch cycles that arrive faster than the monthly rhythm, and for the work to be dominated by finding what runs where, not by installing the fix. Read our Java4Shell story for the same problem from the Java side.

#What to do today

  1. List every host and container image that runs NGINX, and note whether it is the open-source build, NGINX Plus, or a copy bundled in another product.
  2. For each, run nginx -V and check for --with-http_v2_module, then search the configuration with nginx -T for http2.
  3. Upgrade open-source builds to 1.31.9 or 1.30.7, or install your distribution’s fixed package, and restart rather than reload.
  4. Where you cannot upgrade, turn HTTP/2 off on internet-facing listeners as above, and check the three configuration-specific flaws separately.
  5. Ask suppliers of appliances and ingress products the question above, and diarise their answers.

#What we do not know yet

F5 says it has seen no exploitation, and we have not seen any report that contradicts that. We do not know how many internet-facing servers are really vulnerable once backports and configuration are counted, when each distribution will ship its fix, or how long the public proof of concept will take to be turned into something usable against default installations. This story relies on the project’s advisory page and the researchers’ write-ups, not on F5’s own notices on myF5, so check those for the wording that applies to NGINX Plus, and we will update this story if F5’s wording differs.

Evidence trail

Sources and verification

Primary documentation and external technical references used in this article.

  1. 01nginx.org security advisoriesnginx.org
  2. 02Wiz Research: Unauthenticated Remote Code Execution in NGINXwiz.io
  3. 03Xint: NGINX2Root, new remote code execution vulnerabilities in nginxxint.io
  4. 04The Register: AI found more critical bugs in another widely used technology running 225M websitestheregister.com
  5. 05CISA Known Exploited Vulnerabilities cataloguecisa.gov

About this page

Credited to

Dates

Published

Corrections

No corrections are listed for this page. Some review sweeps corrected pages without naming each one; those are in the corrections log.

Reader Interaction

Comments

Add a thoughtful note on F5 Fixes Two NGINX HTTP/2 Flaws That Let an Unauthenticated Attacker Run Code, One Dating Back to 2015. Comments are checked for spam and held for moderation before appearing.

Loading comments...

More news

All news

Discover more

Lexicon Definitions

Was this useful?

Reports go to the site owner and are used to correct pages. Corrections are listed at /corrections.

Get new fixes by email.

New runbooks, configuration traps and fixes from KBY Technologies, about twice a week.