NGINX's maintainer shipped an emergency security release on 18 November fixing two critical flaws in how the web server handles HTTP/2 connections. Tracked as CVE-2026-123007 and CVE-2026-123633, the pair let an attacker who has never logged in corrupt server memory and, in a worst case, run their own code on the machine — no backend application or plugin required. Researchers at Rapid7 and the cloud-security firm Wiz found and reported the bugs separately; one of them, an integer underflow in NGINX's flow-control accounting, has quietly sat in the code since 2015. Wiz demonstrated working code execution against a stock NGINX server doing nothing more than serving static files. Both carry a CVSS score around 9.2, critical. The one piece of good news: pulling this off over a real network means pushing roughly 128GB of HTTP/2 traffic through a connection fast enough to finish within NGINX's one-hour limit, so a practical attack needs serious bandwidth and NGINX configured with HTTP/2 switched on. There are no reports yet of anyone exploiting it in the wild — this is a "patch before someone builds the easy version" warning rather than an active attack in progress. Fixed versions are NGINX Open Source 1.30.7 and 1.31.9, with matching NGINX Plus releases.
What this means for your business
NGINX quietly sits in front of a huge share of the internet — sometimes as the actual web server, more often as the reverse proxy or load balancer that routes traffic to whatever is actually running behind it — so there's a good chance it's somewhere in your stack even if nobody on your team has ever typed its name. Two things worth doing this week. First, ask your developer or hosting provider whether any server they manage runs NGINX, and whether it has been updated to 1.30.7 or 1.31.9 (or the matching NGINX Plus release) since 18 November. Second, if a system can't be patched right away, ask whether HTTP/2 can be temporarily switched off as a stopgap — that alone closes this particular door until the proper update goes out. None of this requires understanding memory corruption or flow control; it just means getting a straight "yes, we're patched" from whoever runs your infrastructure.