Global Edition
Global Edition
UK Edition
EU Edition
US Edition

Understand the story, not the spin.

Markets

Google finds AI framework automating patch-based exploits

Published 16 November 2026 · All stories from this day

Google’s Threat Intelligence Group says it found an artificial intelligence framework that automatically analyzed software patches, generated exploits and attempted to run them, marking the first time the group has observed attackers automating vulnerability discovery and exploitation without human direction.

The framework, called exploit_agent, was found on an exposed staging server, according to Google’s latest AI Threat Tracker. It compared software code before and after security patches to identify likely flaws, then drafted and executed exploits. The system was connected to several language models, including frontier and locally run models.

Google said data on the server suggested the operator targeted French websites in media, entertainment and education. The server also contained stolen victim data and scripts for password brute-forcing, installing webshells and removing webshells left by other attackers. Russian-language comments appeared in the framework’s configuration files, but the language does not establish the operator’s nationality.

The investigation does not establish that exploit_agent’s generated exploits successfully compromised systems, and Google did not...

Now playing
0:00 / 0:00