What matters in AI.

Subscribe

Google finds an AI agent that made attacks from patches with no human

Google tells that this is the first AI agent that it has seen turn patches into exploits that work with no human help.

Google Threat Intelligence Group tells that an operator that it thinks speaks Russian used an AI agent to turn patches into exploits that work. No human was in the loop. Google wrote this in its report for the fourth-quarter of 2026. The server held data from victim companies, but Google gave no count of them.

How it works

Google found a test server that was open to all. It held a framework with the name exploit_agent. Many AI agents work together in it, and it uses the newest models. It compares code before and after a patch, finds weak points, writes exploits and uses them with no human.

What was found

The server also held scripts for use after a break-in. The scripts do scans of the inner network, try passwords many times and install webshells. A webshell is a small tool that gives control of a system from a different place. Google tells that the operator wanted French media, entertainment and education websites.

What is not known

Google gave no test results. It tells that other groups use equivalent tools: CAIRN, CyberStrikeAI, HexStrike AI, and PentAGI. Some are groups that Google connects to China. Google thinks that more groups will use these tools.

Insight: A patch gives attackers a start signal

On 2 November 2026, Google told that ARTEX users went from a first account to deep access in about four hours. Symantec found an agent made to do 100,000 attack steps with no approval. A patch shows attackers where the weak point is, and maybe each patch release is the start for them. Each day without an install gives them more time. Count the days between a patch and your install.

Sources

  1. Google: AI agent turned a security patch into a working exploit with no human helpThe Register
AI MATTER · NEWS · AI MATTER · NEWS ·16 NOV2026

Posted

Tags

Learn the terms in this story

Guides for this story

More in Security

All Security news