The short version
- The flaws. F5 patched 22 vulnerabilities, including five critical NGINX bugs tracked as NGINX2Root with a severity score of 9.2. Successful exploitation lets an attacker run commands with the privileges of the nginx worker process and compromise the server.
- The fixes. F5 resolved the NGINX flaws in NGINX Open Source 1.30.7 and 1.31.9, and in NGINX Plus 37.0.8.1 and 37.3.0.1. The same notification covers 12 BIG-IP bugs and five other NGINX issues.
- Exploit code. Rapid7 and Censys have both published working proof-of-concept code for CVE-2026-123007. Rapid7 says exploitation needs roughly 128 GB of HTTP/2 traffic; Censys says it reproduced the bug with cleartext HTTP/2 and far less traffic.
- Disagreement. F5 says code execution is only possible if address space layout randomisation (ASLR) is disabled or bypassed. Wiz says CVE-2026-123007 and CVE-2026-123633 bypass it themselves, and Rapid7's exploit leaks memory addresses from server responses. F5 makes no mention of any flaw being exploited in the wild.
- Exposure. Censys estimates about 9 million NGINX services run an affected version with HTTP/2 enabled, and says a further 103 million do not reveal their version.
Timeline
| Date | Event |
|---|---|
| 2015 | NGINX 1.9.5 is released; Wiz says every NGINX iteration since is affected by CVE-2026-123007. |
| 19 Nov | F5's patches and public proof-of-concept code for NGINX2Root are reported, with F5 making no mention of any of the vulnerabilities being exploited in the wild. |
Which configurations are affected
F5 issued the fixes in an out-of-band security notification. The five critical NGINX flaws reach code execution through different configurations: CVE-2026-123007 and CVE-2026-123633 need HTTP/2 enabled and lead to a buffer overflow and a use-after-free respectively; CVE-2026-121810 needs HTTP/3; CVE-2026-113425 affects instances acting as mail proxy servers; and CVE-2026-123634 requires a proxy whose upstream server name contains variables, with a subrequest arriving from a Server Side Includes or mirror directive.
A use-after-free is a memory handling bug where a program keeps using a block of memory after releasing it, which an attacker can shape to run their own code. Two of the flaws were independently reported by Ikotas Labs, Metnir, Rapid7, Wiz and Xint. The same notification covers 12 BIG-IP bugs that can lead to remote code execution, denial of service, file manipulation and token lifetime manipulation, plus five NGINX SSL/TLS and denial-of-service issues.
How practical exploitation is
F5 says the critical NGINX bugs can only be exploited if ASLR, the operating system feature that randomises memory addresses, is disabled or can be circumvented. Researchers dispute that constraint. Wiz says two of the flaws can bypass ASLR on their own, and Rapid7's published exploit defeats the protection by leaking memory addresses from server responses.
Rapid7 warns that its proof-of-concept needs about 128 GB of HTTP/2 data to succeed, taking between two minutes and an hour depending on bandwidth; on slow links NGINX's default one-hour connection lifetime prevents it, and a proxied connection may break the attack. Censys says it reproduced CVE-2026-123007 in its lab with cleartext HTTP/2 without generating that volume of traffic, and put its code on GitHub.
Wiz reports that the flaw works against servers doing nothing more than serving static files: "Our testing of CVE-2026-123007 demonstrated code execution on NGINX serving ordinary static files, without a backend application or third-party modules." Xint says it chained the same flaw with a Linux kernel vulnerability to escape a Docker container and gain root on the host.
Scale of exposure
Censys's count is the only estimate in the report of how many systems are reachable. It separates servers that openly run an affected version with HTTP/2 enabled from a much larger group that hides its version, so the true number of vulnerable instances cannot be read from either figure. F5 has not said how many deployments are affected, and has not reported any exploitation in the wild.
What is not known yet
- Whether any of the five NGINX flaws has been used against a real target; F5 reports no in-the-wild exploitation.
- How many of the roughly 103 million NGINX services that conceal their version are running an affected release with HTTP/2 enabled.
- Whether Censys's lower-traffic reproduction holds in production networks where HTTP/2 connections pass through proxies.
- Which of the 12 BIG-IP bugs can be exploited without authentication, and whether any are reachable from the internet.
What to do
If you run NGINX
- Update NGINX Open Source to 1.30.7, or NGINX Plus to 37.3.0.1, following F5's out-of-band notification for the full list.
- Disable HTTP/2, HTTP/3 and mail proxy features on internet-facing servers you cannot patch straight away.
- Check proxies whose upstream server name contains variables and which accept Server Side Includes or mirror subrequests, and remove or restrict those settings until patched.
- Test the Rapid7 and Censys proof-of-concept code against a staging instance to confirm your build is no longer vulnerable.
If you run F5 BIG-IP
- Apply the 12 BIG-IP fixes listed in the same notification.
- Restrict management access to BIG-IP interfaces to trusted networks while patching.
For everyone else
- Ask hosting providers, load-balancer vendors and content delivery networks which NGINX version they run and whether HTTP/2 is enabled on your traffic.
- Inventory internet-facing services that strip version headers, since they cannot be checked by banner alone.
Update log
- First published.
Sources
New incident reports by email
One email a week with the reports published since the last one.
We email you about this topic only; unsubscribe any time. See our Privacy Policy.