Two newly discovered critical flaws in the NGINX traffic routing system allow threat actors to run code on affected servers that have HTTP/2 enabled, all without needing to log in. This could potentially affect millions of websites and put sensitive data at risk.
The vulnerabilities were discussed in a report published by security firm Wiz on Nov. 18, which said the flaws were also independently uncovered by Ikotas Labs, Metnir, Rapid7 and Xint. Both flaws permit unauthenticated remote code execution (RCE).
Developed by F5, NGINX is one of the most popular web server frameworks on the internet and powers some 225 million websites, according to Netcraft research. That means a potentially vast number of sites might end up being compromised if they are not promptly patched.
If successfully exploited, the flaws could allow a malicious actor to hijack NGINX servers, exfiltrate data and disrupt or modify websites. However, a key wrinkle is that NGINX servers must have HTTP/2 enabled in order to be affected.
The vulnerabilities have been tagged as CVE-2026-123633 and CVE-2026-123007, and both carry a severity rating of Critical and a CVSS score of 9.2. That places them comfortably within the most serious classification bracket.
Because the weaknesses lie within NGINX itself, Wiz’s report warned that “Websites and APIs can be exposed even when their own application code is secure.” NGINX deployments often serve multiple applications at once, so a single breach could affect several websites and APIs.
AI uncovered flaws hidden for years
Wiz disclosed that it used Wiz Atlas, its AI code scanning tool, to uncover the bugs, then harnessed Gemini 4 Carbon and ChatGPT 5.6 Sol to validate its findings.
The process demonstrated how AI is increasingly being used by defenders to find and patch code weaknesses before they can be abused. Wiz did not state that it had observed either flaw being exploited in the wild.
The integer underflow bug at the heart of CVE-2026-123007 can threaten web servers running NGINX versions dating back to 1.9.5, which was released over a decade ago in 2015.
The use-after-free vulnerability in CVE-2026-123633, meanwhile, had been hiding in the NGINX code for nearly six years. Wiz described these weaknesses as “memory-safety flaws that have historically been extremely difficult to discover and exploit.”
Without AI, they might never have been found, which is a boon for defenders and attackers alike.
F5 released security patches for both CVE-2026-123633 and CVE-2026-123007 on Nov. 18, 2026, and organizations are strongly recommended to check if their platforms are affected and to install the patches if they are. If that isn’t immediately possible, Wiz advised users to disable HTTP/2 as a temporary workaround.