Back to News
News · AI & Security

Google Finds an AI Framework That Turns Patches Into Exploits, With No Human Involved

Google's Threat Intelligence Group says attackers built a framework that diffs a software patch, finds the underlying flaw, and writes and runs a working exploit on its own.

Source: VentureBeat · November 16, 2026

VentureBeat reports that Google's Threat Intelligence Group has disclosed an AI framework, called exploit_agent, built by attackers to compare code before and after a software patch, identify the vulnerability it fixes, and then draft and run a working exploit, all without a human in the loop. Google says this is the first time it has seen attackers use an LLM-based framework to fully automate both discovery and exploitation. Researchers found the framework on an exposed staging server alongside stolen victim data and post-compromise scripts, though the report stops short of tying the framework itself to any specific breach.

The report also leaves gaps: Google did not name the vulnerability exploited, estimate a victim count, or say how quickly the framework turns a patch into a working exploit. Its configuration contains Russian-language comments, but Google did not attribute it to any group or country, and telemetry suggested an interest in French media, entertainment, and education sites. Google published a code excerpt showing the framework examining access controls in a WordPress plugin, without confirming a successful exploit resulted.

Why it matters: diffing a patch to find the flaw it fixes used to take a skilled human days. If attackers can now automate that step end to end, the window between a vendor's patch and a working exploit keeps shrinking, raising the bar on how fast enterprises must apply patches and monitor their own exposed tooling and staging infrastructure.