· via Hacker News – Front Page (native)
NGINX patches two critical pre-auth HTTP/2 remote code execution flaws
Two unauthenticated RCE vulnerabilities in NGINX's HTTP/2 handling, both scoring 9.2 on CVSS 4.0, can expose sites even when application code is secure; F5 shipped fixes on November 18.

Two unauthenticated remote code execution vulnerabilities in NGINX's HTTP/2 handling were disclosed on November 18 alongside a security release from F5, the company that maintains the web server. Tracked as CVE-2026-123633 and CVE-2026-123007 and both rated critical with a CVSS 4.0 score of 9.2, the flaws could let an attacker run arbitrary code on an affected server without any credentials.
According to Wiz, whose research team found the bugs alongside Ikotas Labs, Metnir, Rapid7 and Xint working independently, the problems sit in NGINX itself rather than in the applications it fronts. A website or API with otherwise secure code can still be exposed, since NGINX commonly terminates TLS and routes traffic for multiple upstream services from a single shared entry point.
How the bugs work
CVE-2026-123633 is a use-after-free: memory tied to an HTTP/2 connection is released while a request still depends on it. Wiz says it manifests in reverse-proxy processing when available connection slots run low. The bug has been present for nearly six years.
CVE-2026-123007 is an integer underflow in HTTP/2 flow-control accounting, where an incorrect send_window value leads to an invalid memory copy. This flaw reaches back to NGINX 1.9.5 from 2015, meaning it has lurked in the code for more than a decade. Wiz demonstrated code execution with it against a server serving ordinary static files over HTTPS on a 64-bit system, with no backend application or third-party modules involved.
Both vulnerabilities require HTTP/2 to be enabled on the affected server. Wiz also notes that typical memory-safety mitigations such as ASLR do not help here, because either flaw can be leveraged to read memory before achieving code execution, defeating address randomization.
Disclosure timeline
Wiz reported the use-after-free to F5 on October 21, 2026, including a confirmed exploit with an ASLR bypass and a proposed fix. F5 confirmed that another research team had previously reported the same issue, and supplied a patch for review on November 10. Wiz reported the second bug on November 16, F5 matched it to CVE-2026-123007 the following day, and the coordinated security release landed on November 18.
AI-assisted discovery
Wiz says it found both bugs using Wiz Atlas, the engine behind its AI code scanning product, and validated their impact using Gemini 4 Carbon and ChatGPT 5.6 Sol. The company frames the findings alongside its recent PHP2Shell work and the Java4Shell disclosure by Calif as evidence that memory-safety bugs once considered too intricate to find reliably are now within practical reach of both researchers and attackers, and that complexity alone can no longer be relied on to keep old flaws hidden. Multiple independent teams converged on the same NGINX bugs after years of obscurity, which Wiz calls out as a particularly striking signal of that shift.
What operators should do
Wiz's guidance is direct: update affected deployments to the latest release, which bundles multiple security fixes. Where an immediate upgrade is not possible, temporarily disabling HTTP/2 removes the precondition both exploits depend on. The company also recommends inventorying deployments to identify instances running affected versions and configurations, prioritizing those reachable from the internet, and applying mitigations while patches are tested and rolled out.
Why it matters
NGINX fronts an enormous share of the public web — Wiz puts the figure at more than 225 million websites — so pre-auth RCE at that layer bypasses the security of everything behind it. A single compromised proxy can expose TLS-terminated traffic, sensitive data and multiple downstream applications at once, which is why these CVEs deserve urgent attention even for teams whose own application code is clean. The disclosure also points to a changing threat landscape: if AI tooling can surface decade-old memory-safety bugs quickly, defenders should assume the window between publication and active exploitation will keep shrinking, and patch velocity becomes the primary defense.
- #nginx
- #security
- #http2
- #vulnerabilities
- #web-server