
Researchers have discovered a ten-year-old flaw in web server software NGINX, marking the latest in a string of critical vulnerabilities in foundational infrastructure detected using AI tools.
Classified as CVE-2026-123007, the flaw is colloquially as ‘nginx2root’. Discovered independently by multiple research teams including Rapid7 and Wiz Research, it was first identified by AI vulnerability discovery platform Xint.
The flaw itself allows attackers to carry out remote code execution or crash servers and stems from an integer underflow in NGINX’s HTTP/2 support. Attackers have not been spotted exploiting it in the wild, and the requirements to carry out an attack are high, but the severity has been rated 9.2 out of 10 by NGINX owner F5.
Nginx2root is the latest example of researchers discovering a vulnerability in widely-used software that had gone undetected for years, with researchers noting that the flaw has been present in NGINX since 2015’s version 1.9.5 release.
It follows flaws including PHP2shell which was present for almost 15 years, as well as Java4shell, which was first introduced in January 2023.
“Seeing two serious security flaws made public a day apart is a major red flag for anybody who hasn’t recognised that the rules of the game have changed,” said Roshan Piyush, Senior Principal Security Research Engineer at AI development company Harness.
“Java4Shell and nginx2root aren’t obscure problems with a niche impact. Both give attackers the ability to take control of a server and sit in software that runs a huge part of the internet. These tools work quietly in the background of banks, airlines, and telecom networks – integrated into the online services that we all rely on every day. Most people never think about them, but they have the potential to cause chaos.”
Sign up today and you will receive a free copy of our Future Focus 2026 report - the leading resource for IT decision-maker insight on priorities and investment areas in AI, security and more.
These instances are part of a growing trend that is already causing severe disruption within the security community.
As AI tools become more sophisticated, they are increasingly being used to audit code, which has resulted in a sharp uptick in the number of bugs being identified, many of which have been hiding in established codebases for many years.
A recent report from Microsoft security researchers, for example, used a pre-release code-audit product codenamed ‘MDASH’ to scan a subset of Linux kernel subsystems. The process – which took one researcher and one intern just three months to complete – resulted in 59 upstream bug reports and 33 CVEs.
Of those 59 bugs, the average age of the flaw was over eight years, with three vulnerabilities being over 20 years old.
The avalanche of new security flaws being discovered and disclosed has led to a dramatic gear-change within the industry.
Software vendors and open source maintainers have had to step up the speed at which they’re able to patch these flaws, while IT leaders now have orders of magnitude more patches to apply within their environments.
Knock-on effects
This phenomenon is also having other knock-on effects; the latest version of database software PostgreSQL was originally scheduled to release on 29 October, but was pushed back to 19 November to accommodate the number of bug fixes, which jumped by 86% compared to the previous version.
“Bugs that have lain dormant for years – over a decade in the case of nginx2root – can now be found in hours, with just a few hundred dollars’ worth of AI tokens,” added Piyush. “That’s a massive problem for companies that still rely on security playbooks that were written even a year ago.”
“It’s just not possible anymore to patch everything by hand and do it quickly enough to prevent an exploit. Security teams have to move at the same pace as attackers, using AI to automatically block vulnerable builds from shipping, identify those already in production, and rollout fixes that keep systems secure while patches are implemented.”