F5 released Nginx 1.31.9 and 1.30.7 to fix two HTTP/2 vulnerabilities as well as eight other flaws, several discovered with AI agents. The most serious allows remote code execution without authentication. The American agency CISA has not yet noted any exploitation of these two vulnerabilities in the wild.
Verify that fixes are running
A version number is not enough to establish that a server is protected, warns Daniel Shechter, co-founder and CEO of Miggo Security: some distributions integrate corrections by backport without changing this number. A simple reload doesn’t replace the executed code either, he points out. In his eyes, the main risk lies with the teams who believe they are wrongly protected, more than the 11 years during which the bug went unnoticed. It recommends considering the server exposed until the patch has actually been executed.
For Nginx Plus, the recommended versions are 37.3.0.1 on the new branch and 37.0.8.1 for the 37.0 branch. Branches R33 to R36, now out of support, will remain uncorrected. F5 recommends an immediate refresh or shutdown of HTTP/2 on the affected ports. This last measure blocks the two most severe vulnerabilities, but notably interrupts native gRPC.
It does not neutralize CVE-2026-123634, CVE-2026-121810 and CVE-2026-113425, Xint’s other discoveries, which require special configurations. The resolver vulnerability affects installations choosing their destination server for each request with the mirror directive; that of HTTP/3 concerns QUIC listeners; that of the mail proxy involves auth_http. Xint has released mitigation measures for each.
Wiz advises examining abnormal terminations of Nginx worker processes as well as the commands they issue. Daniel Shechter also warns that the HTTP/2 traffic used by the attack escapes WAFs and access logs.
Functional exploits under certain conditions
Shodan counts 56.7 million Nginx hosts; of those who expose their version, 98.9% fall into the vulnerable range. All research teams produced a working exploit, and two rated exploitation as easy. Xint claims that an externally reachable TLS/HTTP/2 port is enough to corrupt the send window, reveal addresses and then execute commands. The company says it validated its attack on public IPs at AWS, GCP and Azure, in the presence of competing traffic.
At Metnir, exploiting CVE-2026-123633 took around 10 minutes with ASLR enabled, with most of the time used to bypass this randomization of memory addresses. The team claims to have replicated the freed memory usage in OpenResty and caused crashes in distributed Kong, Apache APISIX, NGINX Gateway Fabric, and community ingress-nginx binaries. These software enable HTTP/2 by default.
The evening letterFree · 7 p.m.
Applied AI, concretely?
Use cases that really work, deciphered every evening in 5 minutes. Free.
Included upon registration: our selection of the best AI guides & comparisons.
Choose your pace
Free · No spam · Unsubscribe in 1 click
Stephen Fewer, principal security researcher at Rapid7, puts the volume needed for his exploit at around 128 GB of HTTP/2 data. He estimates their transmission at around two minutes with 10 Gbit/s, compared to 17 minutes with 1 Gbit/s. Its technique also depends on a glibc copy routine reserved for processors with AVX-512, common extensions in servers.
Yordan Ganchev, senior threat intelligence specialist at watchTowr, however, tempers the announced ease: according to him, the target must be directly accessible, without WAF or CDN ensuring HTTP/2 termination. You also need to organize the memory heap and adapt the exploit to Nginx versions, compilations and configurations. On a server exposed under load, he warns, attempts can result in more crashes than successful code executions.
Multiple teams discover the same flaws
Nginx, whose share is estimated at a third of the Internet, presents with CVE-2026-123007 a flaw likely to threaten millions of sites. The mechanism is based on the manipulation of windows regulating HTTP/2 flows, to the point of causing an out-of-bounds memory copy. Xint.io claims to have obtained by this means the execution of commands with the rights of an Nginx worker process. All official versions since 1.9.5, which introduced HTTP/2 in 2015, are affected, i.e. more than ten years.
Rapid7, Wiz, Xint and Ikotas Labs reported this vulnerability at F5. The other HTTP/2 flaw, CVE-2026-123633, called Drainbreak, corresponds to a use-after-free in versions 1.19.7 to 1.31.8. Metnir, Wiz and Xint discovered it and then reported it.
These findings overlap despite independent research using AI. Stephen Fewer does not mention this usage in his report; the other four teams employed AI agents or platforms. Xint claims five discoveries thanks to its autonomous AI, and its CEO Brian Pak says the platform scanned 50,000 rows in 90 minutes to identify a critical vulnerability allowing remote execution without authentication. Other vulnerabilities have recently been spotted with AI in PHP, Node, OpenSSH and Java.
Related Linux flaw remains unpatched
Xint also claims to have discovered a vulnerability in SCTP, a subsystem of the Linux kernel, which can be combined with an Nginx flaw to gain root privileges. The company called this set nginx2root. During its testing, it says it broke through the isolation of a Docker container running Nginx, then gained root rights on the host machine.
The vulnerability has existed since Linux 4.15, released in 2018, and does not yet have a CVE or fix. The company is keeping its details confidential until distributions can release a fix. In the meantime, administrators not using SCTP can prevent SCTP from loading by writing “install sctp /bin/false” to a file in the /etc/modprobe.d directory.

