AI unearths ‘internet breaker’ flaws in Nginx web server F5 shipped Nginx 1.31.9 and 1.30.7 to fix a critical unauthenticated remote code execution flaw, CVE-2026-123007, that has existed in every official Nginx release since version 1.9.5 added HTTP/2 support in 2015. Xint.io researchers, using the company's autonomous AI harness, found five of the flaws and chained them with an unpatched Linux SCTP kernel bug (present since Linux 4.15 in 2018) to escape a Docker container and gain root on the host; Xint CEO Brian Pak said the platform scanned "50,000 lines of code in just 90 minutes." Shodan shows 56.7 million hosts running Nginx, and 98.9% of those disclosing a version report one in the vulnerable range, though CISA says neither CVE-2026-123007 nor the use-after-free CVE-2026-123633 has been exploited in the wild. Nginx, the web server estimated to power a third of the internet, has shipped fixes for multiple vulnerabilities, including a critical unauthenticated remote code execution flaw that puts millions of websites at risk and has existed in the code for over a decade. Researchers from several security firms independently found some of the same flaws by using AI agents, adding to a string of recent AI-discovered vulnerabilities in widely used internet technologies such as PHP, Node, OpenSSH, and Java. The most severe flaw, tracked as CVE-2026-123007, lets an attacker manipulate flow-control windows in the HTTP/2 protocol implementation to trigger an out-of-bounds memory copy that researchers from Xint.io turned into command execution with Nginx worker privileges https://xint.io/blog/nginx2root-remote-code-execution . The vulnerability affects every official Nginx release for more than 10 years since version 1.9.5, which originally added HTTP/2 support in 2015. A second HTTP/2 flaw, CVE-2026-123633, is a use-after-free vulnerability affecting releases 1.19.7 through 1.31.8. Both are fixed in Nginx 1.31.9 and 1.30.7, which F5 released https://nginx.org/en/security advisories.html along with fixes for eight other vulnerabilities. Xint’s researchers, who used the company’s own autonomous AI harness, found five of the flaws and discovered a separate Linux kernel vulnerability in the SCTP subsystem that can be chained with any of those flaws to gain root privileges on the server. The company has dubbed the collection nginx2root https://nginx2root.com/ . In their tests the exploit chain escaped a Docker container running Nginx and obtained root on the underlying host. The kernel flaw has existed since Linux 4.15, released in 2018, and has no CVE or patch so the company is withholding details until distributions can ship a fix. The exposure of web servers on the internet is vast with the Shodan search engine showing 56.7 million hosts running Nginx, and 98.9% of those that disclose their version report one in the vulnerable range. F5 recommends updating immediately or disabling HTTP/2 on affected ports, which blocks the two most severe flaws but not Xint’s other findings — CVE-2026-123634, CVE-2026-121810, and CVE-2026-113425 — which are in other components. The CVE-2026-123007 flaw was reported to F5 by researchers from Rapid7 https://www.rapid7.com/blog/post/ve-cve-2026-123007-critical-nginx-http-2-integer-underflow-leads-to-rce-fixed/ , Wiz https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities , Xint, and Ikotas Labs https://x.com/ikotas00/status/2123071157083754677 , while CVE-2026-123633, dubbed Drainbreak, was discovered and reported by researchers from Metnir https://metnir.com/research/drainbreak-nginx-http2 , Wiz, and Xint. Only Rapid7 senior principal security researcher Stephen Fewer doesn’t mention AI in his write-up. The other four teams used AI agents and harnesses, with Xint’s CEO Brian Pak saying that his team’s platform scanned “50,000 lines of code in just 90 minutes” to surface “a critical vulnerability that gave an unauthenticated user a RCE path — essentially an internet breaker.” No one has exploited either flaw in the wild yet, according to the US Cybersecurity and Infrastructure Security Agency’s assessment of both CVEs, but every research team built working exploits, and two describe the flaws as readily exploitable. Xint called CVE-2026-123007 highly exploitable, saying an externally reachable TLS/HTTP/2 port was enough to corrupt the send window, disclose addresses, and execute commands, and validated the attack over public IPs on AWS, GCP, and Azure under concurrent traffic. Metnir ran its exploit for CVE-2026-123633 with ASLR enabled, taking about 10 minutes, nearly all of it spent on defeating memory layout randomization, and reproduced the use-after-free in OpenResty, crashing the shipped binaries of Kong, Apache APISIX, NGINX Gateway Fabric, and community ingress-nginx, all of which enable HTTP/2 by default. Rapid7’s Fewer calculated that running his exploit requires transmitting roughly 128GB of HTTP/2 data to the target. That takes about two minutes on a 10Gbit/s network and 17 minutes on 1Gbit/s. The technique also relies on a memory-copy routine in glibc that is used only on CPUs with AVX-512 extensions, but these are common on servers. “Exploitation is by no means a walk in the park, as the vulnerable systems must be directly exposed to the internet with no WAF or CDN terminating HTTP/2,” Yordan Ganchev, principal threat intelligence specialist at watchTowr, told CSO. “Memory corruption can also be tricky, as an effective exploit must properly groom and shape the memory heap and handle different NGINX versions, builds, and configuration variants. And when the system is internet-facing and under load, exploitation can cause more crashes than successful code execution.” Nginx Plus users should also update to 37.3.0.1 on the newer release line or 37.0.8.1 on the 37.0 line. The older R33 through R36 release line will never receive a fix because those versions have reached the end of technical support. Disabling HTTP/2 comes with costs, as services that depend on it, such as native gRPC, stop working. Xint’s other three flaws require specific configurations. The resolver flaw affects setups that pick their backend per request and use the mirror directive, the HTTP/3 flaw affects QUIC listeners, and the mail proxy flaw affects deployments that use proxy on with auth http . Xint published per-flaw mitigations in its technical write-up. For the accompanying Linux flaw, until a kernel patch is released, administrators who don’t use SCTP can block the module from loading by adding install sctp /bin/false to a file under /etc/modprobe.d . Administrators should also investigate unexpected Nginx worker crashes or commands launched by Nginx workers, Wiz noted. “The real risk with this nginx bug isn’t the 11 years it went unnoticed, it’s how many teams will think they’ve patched when they haven’t,” Daniel Shechter, co-founder and CEO of Miggo Security, told CSO. “The version number can be wrong because distros backport fixes without changing it, a reload doesn’t swap out the running code, and the attack happens in HTTP/2 traffic that WAFs and access logs never see. If you can’t prove the fix is actually running, assume you’re still exposed.”