Xint AI Finds Five Nginx RCE Bugs; F5 Patches Wednesday
TL;DR
- Xint says its OpenAI-powered harness found four of five Nginx RCE flaws in 50,000 lines of source in 90 minutes without human guidance.
- F5 patched the five 'nginx2root' bugs in Nginx 1.30.7 and 1.31.9 on Wednesday; all five carry a CVSS 4.0 score of 9.2.
- Wiz, Rapid7, Ikotas Labs and Metnir independently reported the two HTTP/2 flaws, CVE-2026-123633 and CVE-2026-123007, to F5.
Cybersecurity startup Xint says an OpenAI-powered system it describes as 'the world's best AI hacker' chewed through about 50,000 lines of NGINX source in 90 minutes and returned four of five vulnerabilities with no human guidance, in research Forbes first reported on Wednesday. F5, which maintains NGINX, patched the five flaws, collectively dubbed 'nginx2root,' in versions 1.30.7 and 1.31.9 the same day; all five carry a CVSS 4.0 score of 9.2.
Two of the bugs sit in NGINX's HTTP/2 implementation. CVE-2026-123633 is a use-after-free on HTTP/2 connection reuse, and CVE-2026-123007 is a flow-control accounting error that triggers an invalid memory copy; both enable unauthenticated remote code execution when HTTP/2 is turned on. Xint chained one of them with a flaw in the Linux kernel's SCTP networking code to reach root, the highest privilege level on the host.
'Pre-auth RCE in nginx is the bug every hacker dreams of finding. It almost never happens,' Xint chief executive Brian Pak wrote on X. Xint, which works with the Pentagon's DARPA unit, spent roughly $1,000 in OpenAI tokens on the run, using the GPT 6.1 Sol and GPT-6 Luna models that have been generally available since May.
Xint was not alone. Wiz, Rapid7, Ikotas Labs and Metnir independently reported the same two HTTP/2 flaws to F5; Wiz says its Atlas harness, which powers its AI code-scanning solution, surfaced both. Pak told Forbes his team believes as many as 11.8 million vulnerable NGINX installations are reachable on the public internet, and warned that 'the speed at which AI-enabled attackers can respond and reverse engineer exploits from even the smallest clues means disclosure processes established over decades are changing.'
Mitigation guidance is consistent across the camps: patch to 1.30.7 or 1.31.9, or disable HTTP/2 on exposed listeners, which blocks the two HTTP/2 flaws but not the other three Xint disclosed.
Shared on Bluesky by 1 AI expert
-
🚨BREAKING🚨 AI has generated a "holy grail" hack on another core piece of internet infrastructure. Using OpenAI models, Darpa-contracted hackers Xint found five weaknesses in the NGINX server software. The most serious …
View on Bluesky →
Originally reported by forbes.com
Read the original article →Original headline: Xint Finds Five Nginx HTTP/2 RCE Flaws Using GPT-6.1 Sol, F5 Patches Wednesday