exploit_agent: AI Turns Patches into Attacks
· French media, entertainment and education websites
Google found an attack pipeline that read security patches and fired exploits on its own.
Open on the live 3D map →- Target
- France
- Threat actor
- Suspected Russian-speaking operator (per Google GTIG)
- Attribution
- Non-state / unknown Official attribution
- Impact
- First observed attack chain run end to end by AI, with no human in the loop
What happened
Google’s Threat Intelligence Group and Mandiant found an exposed staging server running a multi-agent framework they call exploit_agent, wired to several frontier and locally hosted language models. It watched popular web applications for new patches, compared the code before and after to work out what had been fixed, drafted an exploit and launched it automatically. Stolen victim data sat on the same server; reports say French sites were among the targets.
Aftermath
GTIG published the case in its November 16 AI Threat Tracker, calling it a significant threshold: in the previous edition it had not yet seen fully autonomous chains against real targets. It also described a China-nexus group building its own automated framework.