ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
NGINX Hit by Two Unauthenticated RCE Flaws: Hidden for Eleven Years, Found by AI

NGINX Hit by Two Unauthenticated RCE Flaws: Hidden for Eleven Years, Found by AI

AI information • Admin • • 14 views

On November 18, Wiz and F5 jointly disclosed two critical vulnerabilities in NGINX: CVE-2026-123633 and CVE-2026-123007, both scoring 9.2 under CVSS 4.0, which let an attacker execute code remotely on a server without any authentication. Both flaws hide in NGINX's HTTP/2 handling, and CVE-2026-123007 traces back to NGINX 1.9.5, released in 2015 — it sat in the code for more than a decade. NGINX is the world's most widely used open-source web server, powering more than 225 million websites, and this time the problem is in NGINX itself rather than the code behind it — meaning that even a perfectly written application can be breached straight through its front door.

Both are memory-safety issues, but they trigger differently. CVE-2026-123633 is a use-after-free: an HTTP/2 connection still processing a request is wrongly marked reusable, then closed and its memory freed when connection slots run low, while request handling is still using that memory. CVE-2026-123007 is an integer underflow in the send window: a flow-control refund does not check the upper limit, and combined with unsigned-integer wrap-around it defeats a bounds check entirely, ending as an out-of-bounds memory copy on the TLS send path. Wiz's testing confirmed remote code execution on an NGINX instance serving only static files, with no backend application at all — and standard mitigations such as ASLR do not stop this kind of attack.

IDFlawFixed releases
CVE-2026-123633HTTP/2 connection use-after-freeOSS 1.30.7 / 1.31.9; Plus 37.0.7.1 / 37.2.0.1
CVE-2026-123007HTTP/2 send-window integer underflowOSS 1.30.7 / 1.31.9; Plus 37.0.8.1 / 37.3.0.1

What really deserves attention in this disclosure is not the flaws themselves, but how they were found. Wiz discovered both with Wiz Atlas, its own AI code-scanning harness, then used Gemini 4 Carbon and ChatGPT 5.6 Sol to confirm they really allow remote code execution. Wiz puts it bluntly in the post: memory-safety flaws like these used to be extremely hard to find, because tracing the subtle interplay between memory operations and multiple code paths took enormous human effort — complexity itself was a barrier, and AI is dissolving it. The same script has now played out three times within a month: first PHP2Shell, then Java4Shell, and now NGINX — and multiple independent teams dug up the same flaws at the same time.

For operations teams, the to-do list is short: first find out which deployments have HTTP/2 enabled, then upgrade to a fixed release as soon as possible; where an upgrade cannot happen immediately, disable HTTP/2 for now and re-enable it after upgrading. Also watch for abnormal NGINX worker crashes and suspicious commands spawned by NGINX processes — both can be signs of active exploitation. The patch is out; what remains is a race against time, because once vulnerability details are public, the window between disclosure and large-scale exploitation only gets shorter.

Recommended Tools

More