AI System Exposes 18-Year-Old NGINX Vulnerability, Forces Global Patching Effort

A critical security flaw discovered by an autonomous AI system threatens millions of web servers today as platform engineers race to patch an 18-year-old vulnerability in NGINX. The flaw, nicknamed “NGINX Rift,” allows for unauthenticated remote code execution or persistent denial-of-service attacks across infrastructure powering everything from simple websites to high-stakes AI backends.

The security researchers at Depthfirst disclosed the vulnerability May 13, 2026, after their autonomous auditing system scanned the open-source web server’s codebase. The AI identified multiple memory corruption issues in six hours of analysis, flagging a critical heap buffer overflow tracked as CVE-2026-42945. The bug resided in the ngx_http_rewrite_module and triggered when a rewrite directive using an unnamed PCRE capture, such as $1, appeared with a question mark in the replacement string and was followed by another rewrite, if, or set directive.

F5, the maintainer of NGINX, confirmed the vulnerability in an official advisory and assigned it a CVSS score of 9.2. The flaw dated back to version 0.6.27, released in 2008, and affected every open-source release through 1.30.0. Impacted commercial versions included NGINX Plus releases R32 through R36. F5 stated that on hosts where Address Space Layout Randomization (ASLR) was disabled, the overflow delivered full unauthenticated remote code execution (RCE). On systems with ASLR enabled, attackers possessed the capability to crash worker processes in a continuous loop, creating a reliable path for denial-of-service attacks.

Depthfirst published a technical breakdown and a working proof-of-concept (PoC) exploit on GitHub within hours of the public disclosure. The PoC demonstrated how the mismatch between NGINX’s buffer size calculation and its subsequent copying of escaped data created the overflow condition. The rapid release of a functional exploit caught the industry flat-footed. Real-time internet scans showed roughly 19 million internet-facing NGINX instances running vulnerable version banners. The United States hosted the highest concentration with more than 5.3 million exposed servers, followed by China at 2.54 million and Germany at 1.87 million.

The NGINX project responded by shipping fixes in stable release 1.30.1 and mainline 1.31.0. NGINX Plus users received updates via R36 P4 and specific backports. For operators unable to patch immediately, F5 recommended a configuration workaround: replace every unnamed capture ($1, $2) with a named one (?<name>) inside affected rewrite rules. Security analysts noted that while the total exposure count was massive, the actual trigger required a specific configuration pattern that generic scanners could not immediately detect.

The discovery carried significant weight in the cybersecurity community because it originated from an autonomous AI system rather than human manual review. Depthfirst reported that its tool surfaced four separate remote memory issues, including the “Rift” flaw that sat latent since the rewrite engine’s early development. The speed of the discovery collapsed the typical timeline between code review and public exploitation, highlighting a new reality for teams managing legacy plumbing. Platform engineers who previously viewed NGINX as “set-it-and-forget-it” infrastructure began auditing rewrite rules globally to close the 18-year-old security gap.

Chain Street’s Take

The NGINX Rift discovery proves that “security through obscurity” is officially dead in the age of autonomous AI auditing. An 18-year-old bug sat in one of the most scrutinized codebases on earth until a machine found it in six hours, immediately putting 19 million servers at risk. For the crypto and AI sectors, where NGINX often serves as the primary ingress for high-value APIs and Kubernetes clusters, the “set-it-and-forget-it” mindset is now a liability. This event signals a shift where legacy infrastructure gets stress-tested at machine scale. Operators must prioritize automated patching cycles because the next ancient flaw is already being indexed by a rival AI scan.