Google says attackers built an AI framework to turn software patches into exploits without human intervention
Key Points:
- Google’s Threat Intelligence Group uncovered an AI framework called exploit_agent that automates turning software patches into exploits without human involvement, marking a first in fully automated vulnerability discovery and exploitation using large language models (LLMs).
- The framework continuously compares pre- and post-patch code to identify vulnerabilities, draft exploits, and execute them autonomously, streamlining a process that previously took skilled humans days to complete.
- Discovered on an exposed staging server alongside stolen data and post-compromise scripts, exploit_agent’s operator appeared to target French media, entertainment, and education websites, though no direct attribution to a group or country was made.
- Google did not specify particular vulnerabilities exploited, victim counts, or patch-to-exploit timelines, but highlighted that this tool represents a significant advancement in attacker automation compared to prior partially automated methods.
- Security experts emphasize the need for faster patching and vigilant monitoring of exposed internet assets to counteract the increased speed and efficiency attackers gain through AI-driven exploitation tools.