A decade-old HTTP/2 flaw in NGINX lets attackers run code without authentication
On November 18, 2026, F5 released NGINX 1.31.9 and 1.30.7 to fix CVE-2026-123007 and CVE-2026-123633, two HTTP/2 memory-corruption bugs found independently by five research teams. Disable HTTP/2 or apply the patch now: a single request is enough to execute code, even on a server serving only static files.
November 18, 2026. F5 releases NGINX 1.31.9 and 1.30.7, two builds that fix CVE-2026-123007 and CVE-2026-123633, a pair of memory-corruption bugs in the HTTP/2 module. November 18, 2026. Researchers at Wiz, Ikotas Labs, Metnir, Rapid7, and Xint disclose that they each found one or both flaws on their own. November 18, 2026. Wiz demonstrates remote code execution against a server serving only static files, with no backend application and no third-party modules. Why it matters: NGINX is the most widely deployed web server on the planet, powering more than 225 million websites, and it frequently sits as the single front door to dozens of internal applications — one compromised instance exposes everything behind it.
A memory-corruption bug in HTTP/2 flow control, hiding since 2015
The headline flaw, CVE-2026-123007, nicknamed nginx2root, is a memory-corruption bug in HTTP/2 flow control. Wiz describes it as an integer underflow: a miscalculated flow-control window value causes a write past the end of a buffer. The decisive detail is its age. It has been in the code since version 1.9.5, released on September 22, 2015 — more than ten years. The second, CVE-2026-123633, is a use-after-free in HTTP/2, present since 1.19.7, roughly six years.
Both flaws live in NGINX itself, not in an extension, a library, or an application. They need no exotic configuration: HTTP/2 simply has to be enabled on the server. And HTTP/2 is the option people have ticked by reflex for years, for latency and multiplexing. The trigger comes down to one condition: code execution assumes ASLR is off, or can be bypassed. Wiz says it bypassed ASLR in its own testing — so the practical barrier has already fallen.
Five teams, one conclusion, and AI dissolving the complexity barrier
The most striking part of this story is not the bug itself, but the number of teams that found it in parallel: Wiz Research, Ikotas Labs, Metnir, Rapid7, and Xint. When five independent labs uncover the same memory bug, buried for years inside the world’s most popular web server, that is not a coincidence. Wiz says it used Wiz Atlas, its code-scanning engine, and then Gemini 4 Carbon and ChatGPT 5.6 Sol to validate the impact of both flaws. The researchers’ argument is clear: the complexity of these memory bugs was, for years, its own protection — tracing the subtle interactions between allocations and code paths took enormous manual effort. AI dissolves that barrier, and every new generation of models brings flaws into reach that were once out of it.
The practical consequences are already documented. Xint says it executed code with nothing more than a reachable HTTP/2 port — no special configuration. And the escalation does not stop at the process: the code gained runs with the privileges of an NGINX worker, but Xint reports reaching root on the host by chaining a separate bug in the Linux kernel’s SCTP code, even when NGINX was running inside a Docker container. At publication time, no in-the-wild exploitation had been confirmed, CISA had not added the flaw to its KEV catalog, and a proof-of-concept was already circulating on GitHub — with a note that it only demonstrates the attack against a freshly started container on the same machine. Do not plan around that absence of exploitation: the technique is public.
The scope: 1.9.5 through 1.31.8, plus NGINX Plus
The scope is enormous. Per the NGINX security advisories, every open-source release from 1.9.5 through 1.31.8 is vulnerable — including the October 14, 2026 security releases 1.31.7 and 1.30.6. NGINX Plus is affected too: the 37.x branches before the fixed builds, plus R34 through R36. Version 1.31.8, released October 21, 2026, is vulnerable.
The fixed versions are 1.31.9 (mainline) and 1.30.7 (stable). For NGINX Plus, builds 37.3.0.1 and 37.0.8.1 fix four of the five flaws, and CVE-2026-123633 had its fix earlier in 37.2.0.1 and 37.0.7.1. The official Docker images for NGINX exposed the 1.31.9 and 1.30.7 tags on release day, Alpine variants included.
The trap to know about is distributions. As of November 18, neither Ubuntu nor Debian had shipped a fixed package: Ubuntu’s CVE page listed nginx vulnerable across its releases from 16.04 to 26.10, and Debian’s security tracker listed it vulnerable in bookworm, trixie, forky, and sid. A distribution package can carry the fix without changing the number that nginx -v reports — compare the full package version against your distribution’s own advisory rather than trusting the version string alone.
Nor is this fix alone. The November 18 release closes other memory-corruption bugs, three of which can also lead to code execution under the same ASLR condition: CVE-2026-123634 in the resolver, CVE-2026-121810 in HTTP/3, and CVE-2026-113425 in the mail proxy module.
Patching and working around
The complete fix is a version upgrade. Move to 1.31.9 (mainline) or 1.30.7 (stable), or to the fixed NGINX Plus builds, then restart: a plain reload does not replace the running executable, and existing connections can linger on old worker processes. Use a full restart or NGINX’s live binary upgrade.
While you wait for your distribution’s package, the direct mitigation is to disable HTTP/2. On NGINX 1.25.1 and later, add http2 off; inside each affected server block and remove http2 from the listen directives. On versions 1.9.5 through 1.25.0, remove http2 from the listen directive — listen 443 ssl http2; becomes listen 443 ssl;.
# NGINX 1.25.1 and later: turn HTTP/2 off per server block
server {
listen 443 ssl;
http2 off;
# ...
} Then confirm the server no longer accepts HTTP/2. The command below should stop printing ALPN protocol: h2.
openssl s_client -connect example.com:443 -alpn h2 </dev/null 2>/dev/null | grep -i alpn Two field precautions. First, test the origin itself, not the CDN: an origin that only accepts HTTP/1.1 cannot be attacked over HTTP/2, even if the CDN in front negotiates HTTP/2 with clients. Second, know that turning off HTTP/2 blocks CVE-2026-123007 and CVE-2026-123633, but not the other three flaws — disable HTTP/3 too if you use it, and follow the resolver and mail-proxy workarounds the vendor published.
The verdict
If you expose an NGINX with HTTP/2 enabled, apply 1.31.9 or 1.30.7 today — or disable HTTP/2 immediately while you wait for your distribution’s package. The risk concentrates on the front server: a single crafted request is enough to run code with a worker’s privileges, and Xint has shown a path to root already exists, container or not. The fix costs an upgrade and a restart; the compromise costs access to everything the front protects.
If you depend on Ubuntu or Debian packages, do not sit still: at announcement time no distribution had shipped a fix, and both the use-after-free and the integer underflow have been in the code for years — their simultaneous discovery by five teams shows the window between disclosure and exploitation is closing faster than the distribution update cycle. nginx2root is a dated, sourced reminder that when the flaw is in the front end itself, the installed-package inventory is no longer enough.