Unauthenticated Remote Code Execution in NGINX: CVE-2026-123633 and CVE-2026-123007 · Nir Ohfeld and Tomer Schwartz, Wiz Research · wiz.io, November 18, 2026
Wiz says its AI harness found two NGINX remote code execution flaws
F5 released fixes today for two HTTP/2 memory-safety flaws in NGINX. Both are rated Critical at CVSS 4.0 9.2 and both allow unauthenticated code execution. CVE-2026-123633 is a use-after-free that is most likely to be exploited in reverse-proxy setups when free connection slots run low. CVE-2026-123007 is a send-window integer underflow that dates to NGINX 1.9.5 in 2015; Wiz says it confirmed code execution with an ASLR bypass on official 1.31.7 and 1.31.8 builds serving ordinary static files. Fixed open-source releases are 1.30.7 stable and 1.31.9 mainline. If you cannot upgrade yet, Wiz's interim mitigation is to turn HTTP/2 off.
Wiz says it found both bugs with Wiz Atlas, "the harness powering Wiz's AI Code scanning solution", and used Gemini 4 Carbon and ChatGPT 5.6 Sol to validate the impact. Ikotas Labs, Metnir, Rapid7 and Xint found the same flaws independently. Wiz reads that collision as a sign that AI has removed the complexity barrier to finding bugs like these; the post does not say how the other four teams worked. Wiz is withholding its exploits, so the discovery method and the exploitation results rest on its own account and on the fixes F5 shipped. The post had 780 points and 327 comments on Hacker News at 16:02 EST.
Why it matters: Wiz expects the window between disclosure and exploitation to shrink as models improve. If you run services behind NGINX, this release is a test of how fast your pipeline can roll out a server patch, and HTTP/2 off is the fallback while it does.