Unauthenticated remote code execution flaws are rare in NGINX, yet multiple security teams just uncovered five critical vulnerabilities that expose millions of internet-facing servers.
Disclosed on Nov. 18 alongside an out-of-band patch from maintainer F5, the bugs carry a 9.2 CVSS 4.0 severity rating and let unauthenticated attackers execute commands with NGINX worker privileges.
Research firm Xint (Theori) found four of the five bugs autonomously in 90 minutes using an AI testing harness and chained them with an unpatched Linux kernel flaw to seize root host access under the moniker "NGINX2Root."
Five critical attack paths
While Wiz Research, Rapid7, Ikotas Labs, and Metnir independently reported subsets of the bugs, Xint documented all five memory-corruption flaws:
CVE | Vulnerable Component | Primary Requirement |
|---|---|---|
CVE-2026-123007 | HTTP/2 flow control (ngx_http_v2_module) | HTTP/2 enabled on a listening port |
CVE-2026-123633 | HTTP/2 connection draining (ngx_http_v2_module) | HTTP/2 enabled with upstream proxying |
CVE-2026-121810 | HTTP/3 QUIC stream draining (ngx_http_v3_module) | HTTP/3 enabled with upstream proxying |
CVE-2026-123634 | DNS resolver waiting list | resolver, variable proxy_pass, and mirror |
CVE-2026-113425 | Mail proxy authentication (ngx_mail_proxy_module) | Mail proxy using proxy on and auth_http |
CVE-2026-123007 poses the widest risk because it has lurked in NGINX since HTTP/2 support launched in version 1.9.5 in 2015 and requires no backend proxying; even servers hosting static files are vulnerable. Shodan data analyzed by Xint showed 98.9% of 11.94 million version-identifiable NGINX hosts running affected releases.
AI speeds up bug discovery—for everyone
Multiple research teams converging on decade-old, heavily audited C code marks a turning point in offensive security.
"When we used Xint to scan NGINX, we didn’t know what to expect, since it’s a mature, widely deployed codebase that’s undergone years of scrutiny," Brian Pak, co-founder and CEO of Xint, told CyberScoop.
"Others have targeted NGINX with AI pentesting, but none found this critical bug, which is a testament to our team and the engineering expertise that’s gone into our AI harnesses."
Defenders cannot rely on manual code audits or historical stability as proof of safety when AI harnesses can uncover complex integer overflows and use-after-free chains in hours, a capability threat actors can access just as easily.
Operational realities and how server operators should respond
No active exploitation has been publicly reported, though researchers have demonstrated that CVE-2026-123007 can be exploited under certain conditions.
Rapid7 researcher Stephen Fewer published a proof of concept and noted that the demonstration depended on a specific system configuration and a large volume of HTTP/2 traffic. Those constraints make exploitation harder in practice, but affected servers still need patching.
Still, server administrators and DevOps teams running NGINX across cloud infrastructure, containers, or on-premises web gateways must act immediately:
- Upgrade and restart: Update NGINX Open Source to 1.30.7 (stable) or 1.31.9 (mainline), and NGINX Plus to 37.0.8.1 or 37.3.0.1. A configuration reload will not replace the vulnerable running binary; operators must restart the service.
- Verify Linux distribution backports: Ubuntu and Debian often backport security patches without altering the version reported by nginx -v. Check full package numbers against vendor advisories.
- Apply targeted mitigations: Disabling HTTP/2 only blocks CVE-2026-123007 and CVE-2026-123633. Operators using HTTP/3, dynamic DNS proxying, or mail proxies must apply configuration-specific workarounds and block the unpatched Linux kernel SCTP module via /etc/modprobe.d/ where unused.
One place teams can miss NGINX is inside software they didn’t install or package themselves. Check Kubernetes ingress controllers, vendor appliances, and third-party container images for bundled copies.
Those deployments may have a different update path from standalone NGINX servers, so confirm who owns each one and which vendor advisory applies.
Keep the timeline of when each exposed instance was still vulnerable. CERT-EU notes that an NGINX worker crash alone is weak evidence of successful code execution.
If a crash occurred during that window, correlate it with process creation logs before drawing conclusions. Closing the exposure and determining whether anyone used it are two separate jobs.
Other news: Wiz launched an AI Security Agent that scans application code and cloud permissions to find AI agents missing from enterprise inventories and flag potential paths to sensitive data.





