In short: Google's threat team found a framework called exploit_agent that compares software before and after a patch, drafts an exploit and runs it automatically. Published 21 Nov 2026 by the AI Decoded desk; news dated 16 Nov 2026.

Google says attackers have used an AI system to find software flaws and exploit them with no person directing the work. Its Threat Intelligence Group (GTIG) calls this a first.

What Google found

In its latest AI Threat Tracker, published 16 November 2026, GTIG describes a multi-agent framework called exploit_agent, found on an exposed staging server. It works with several frontier and locally run AI models.

The tool continuously compares popular web software before and after each security patch, works out what was fixed, drafts an exploit and runs it. The same server held stolen victim data and scripts for follow-up hacking. Victim data pointed to French media, entertainment and education websites. Russian-language comments appeared in the code, but Google did not name a group or country.

What it doesn't say

Google gave no victim count, success rate or measured time from patch to attack, and the report doesn't prove the framework itself caused every compromise found on the server.

Why it matters

The techniques aren't new. What's new is speed: chaining them together can shrink the gap between a patch being published and attackers using it. GTIG also saw open-source AI hacking tools such as CAIRN and PentAGI used in real intrusions.

What to do

Patch quickly, especially public-facing web apps. Give any internal AI agents the fewest permissions possible, and watch for unusual activity soon after big patch releases.

FAQs

What is exploit_agent?

Google's name for an AI framework found on an attacker's server that automatically studies software patches, writes exploits and runs them.

Who was targeted?

Victim data on the server pointed to French media, entertainment and education websites. Google did not publish a victim count.

Sources

Follow AI Decoded on Substack

Weekly plain-English AI news and verified UK events.