Google: AI agent turned a security patch into a working exploit with no human help
Reported by TMCnet, The Register, SiliconANGLE and 9 more outlets12 sources ↓
Google’s Threat Intelligence Group says it found attackers using exploit_agent, a framework that compares software before and after patches, drafts exploits and runs them automatically. It calls this its first observed use of a fully autonomous exploit pipeline against targets in the wild.
Key points
- exploit_agent ran on an exposed staging server and used several frontier and locally hosted language models
- The server also held stolen victim data and scripts for reconnaissance, brute-forcing and installing webshells
- Telemetry pointed to French media, entertainment and education websites as targets
- GTIG says autonomous tools leave forensic traces that can help defenders detect them
Sources
- TMCnetRead ↗Google Caught an AI Turning Patches Into Exploits With No Human in the Loop. The Patch Window Is Now a Scale Problem
- The RegisterRead ↗Google: AI agent turned a security patch into a working exploit with no human help
- SiliconANGLERead ↗Google says an AI framework turned security patches into exploits with no human in the loop
- MezhaRead ↗Google says AI agent turned security patches into working exploits without human help
- news.lavx.huRead ↗Google: AI agent turned a security patch into a working exploit without human help
Details
GTIG found exploit_agent on an exposed staging server. The framework was configured to compare code for popular web applications before and after patch announcements, identify possible vulnerabilities, draft exploits and execute them. The server also contained data taken from victims and scripts for internal reconnaissance, brute-forcing and webshell operations.
Telemetry suggested the operator focused on French websites in media, entertainment and education. Russian-language configuration comments referred to API keys for autonomous exploitation. GTIG said attackers may find patched vulnerabilities more efficient to target than zero-days because there is a finite body of material to analyze.
The report also describes attackers using open-source agentic penetration-testing frameworks to automate parts of live intrusions. GTIG observed CAIRN in an intrusion targeting the U.S. aerospace sector and a customized PentAGI-based platform scanning organizations in Asia between late August and early October 2026. It said these tools can speed up operations but currently leave forensic traffic and other traces that may aid detection.
Written from coverage as of
Images
- SiliconANGLE
- Unite.AI
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh