Google warns that fully automated cyberattacks have arrived and will get worse fast
Reported by Bleeping Computer and The Decoder2 sources ↓
Google's Threat Intelligence Group says it found attackers using an AI framework to discover and exploit software vulnerabilities without human involvement. The system targeted French websites in media, entertainment and education.
Key points
- The framework, called exploit_agent, connects to multiple language models
- It compares code before and after software patches, then develops and runs exploits
- Google found the system on an exposed server containing attack data
- Russian-language configuration comments suggest the operator speaks Russian, The Decoder reports
Sources
Details
Google says this is the first time it has identified threat actors using language-model frameworks to automate both vulnerability discovery and exploitation without a human in the loop. The exposed server hosted a multi-agent system called exploit_agent, which Google says connects to several frontier and locally run language models.
The Decoder reports that the system examined software changes made in patches, identified vulnerabilities and attempted to exploit them. Server data indicated that the operator targeted French websites in media, entertainment and education. Google says the automated attacks prioritize speed over stealth, leaving traces that can help defenders detect them.
Written from coverage as of
Images
- The Decoder
- Bleeping Computer
- Bleeping Computer
- Bleeping Computer
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh