Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

AI-discovered NGINX flaws expose 225 million sites to remote code execution

Five independent RCE flaws in NGINX, each exploitable without authentication when HTTP/2 is enabled.

Five critical remote code execution vulnerabilities in NGINX, the world's most widely used open-source web server, were disclosed on November 18, 2026, with cybersecurity firm Xint reporting it discovered the flaws using OpenAI's GPT-6.1 Sol and GPT-6 Luna models1,3.

The vulnerabilities, collectively dubbed NGINX2Root by Xint, each independently allow an unauthenticated attacker to execute commands on an affected NGINX server with HTTP/2 enabled. Xint also found a separate Linux kernel vulnerability that can be chained with any of the NGINX flaws to gain root privileges on the host.

Scope and severity

Two of the flaws, tracked as CVE-2026-123633 and CVE-2026-123007, carry a critical CVSS 4.0 score of 9.22,4. CVE-2026-123007 affects versions dating back to NGINX 1.9.5, released in 2015. Because NGINX typically sits in front of multiple applications and APIs, handling HTTPS and routing incoming requests, a single vulnerable proxy can put multiple backend services at risk even when their own application code is secure.

Netcraft counted more than 225 million sites using NGINX in its September 2026 web server survey. Successful exploitation could let attackers take control of affected servers, steal sensitive data, tamper with websites, or disrupt services. The disclosed findings show that remote code execution is possible under the reported conditions but do not establish that attackers have used either flaw against production systems.

Wiz Research, Ikotas Labs, Metnir, and Rapid7 independently uncovered the same two CVEs, according to Wiz. Wiz said it reported its findings to F5, the company behind NGINX, and collaborated with F5's security team on remediation and coordinated disclosure.

The AI discovery angle

Xint told Forbes that its AI is finding vulnerabilities across a range of targets, including the NGINX flaws. Forbes described Xint as a DARPA-backed cyber company. Xint said it used GPT-6.1 Sol and GPT-6 Luna in the discovery process.

ANALYSIS The disclosure marks a concrete, named instance of frontier language models being applied to discover critical-severity vulnerabilities in foundational internet infrastructure, rather than in contrived capture-the-flag exercises or narrow bug-bounty targets.

The disclosure arrives amid a broader wave of AI-linked security events. On November 16, Google Threat Intelligence Group reported identifying the first case it had observed of attackers using an AI framework to automate both vulnerability discovery and exploitation without human intervention[1].

Xint recommends immediate patching and provides commands to check for exposure. Wiz recommends updating affected NGINX deployments to the latest release, which includes multiple security fixes.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.