According to the research findings, NGINX2Shell lets unauthenticated attackers execute arbitrary code. It can also potentially cause a denial of service when HTTP/2 is enabled.
The vulnerability affects NGINX Open Source versions 1.9.5 through 1.31.8 under specific conditions. Version 1.9.5 was released in 2015. So the problematic code has been quietly shipping for years while the rest of the internet built on top of it.
NGINX2Shell didn’t arrive alone. A second bug, CVE-2026-123633, was disclosed alongside it. This one is a use-after-free issue in NGINX’s HTTP/2 implementation.
CVE-2026-123633 affects versions 1.19.7 through 1.31.8. It also comes with additional requirements before it can be exploited, which narrows the field somewhat.
Patches, proofs and a crowded discovery
Both vulnerabilities went public on November 18, 2026. F5, which owns NGINX, released patched versions the same day: NGINX Open Source 1.31.9 and 1.30.7.
Ikotas Labs credited OpenAI for its collaboration during both the discovery and disclosure phases in a post on X dated November 18, 2026.
The research findings also show that Ikotas wasn’t the only team digging in this particular spot. Overlapping vulnerabilities were independently identified by several security firms, including Wiz, Rapid7 and Xint.
Rapid7 went a step further and demonstrated a working proof-of-concept exploit against an official NGINX 1.31.8 Docker image. Its attack required approximately 128 GB of traffic under lab conditions.
As of the disclosure date, public proof-of-concept exploits exist. No active exploitation in the wild had been reported.
Why NGINX bugs hit differently
NGINX is a leading web server and reverse proxy. It supports a significant portion of global internet traffic.
A reverse proxy sits in front of other servers, receiving requests and passing them along. That position makes it powerful, and it makes a compromise of the bouncer especially bad news for everyone inside.
What this means
For anyone running NGINX, the immediate action is unglamorous: upgrade to 1.31.9 or 1.30.7. With proof-of-concept code already public, the gap between disclosure and real-world attacks is the main risk.
The research findings note that any delays or failures in upgrades may lead to severe security breaches. Teams that can’t patch right away should at least review whether HTTP/2 is enabled and how exposed their affected instances are.
The bigger story may be how this flaw was found. OpenAI’s involvement, plus the fact that several firms used AI tools and converged on overlapping bugs, points to a shift in vulnerability research.
For AI labs, disclosures like this double as proof of usefulness. Crediting OpenAI in a coordinated disclosure of a 9.2-rated bug is a concrete, verifiable example of AI contributing to security work.