
Google GTIG Reports First Autonomous AI Exploit Framework Used in Real-World Attacks
Google Threat Intelligence Group has identified the first known use of a multi-agent AI framework called exploit_agent that autonomously discovers vulnerabilities, develops exploits, and executes attacks without human intervention. The framework was found on an exposed staging server alongside stolen data and post-compromise scripts, with targeting focused on French media, entertainment, and education sectors. State-sponsored and cybercrime groups are rapidly adopting tools such as Artex, Cairn, CyberStrikeAI, and PentAGI to accelerate n-day exploitation and multi-stage intrusions at machine speed. Chinese APT groups, Iran-linked APT42, and North Korean operators including Midnight Neptune are among those integrating LLMs and agentic frameworks into espionage and cryptocurrency attacks. Google also observed commercial Influence-as-a-Service platforms and LLMJacking operations that monetize stolen credentials, while noting that increased AI adoption introduces forensic noise that can degrade long-term operational security. Experts warn that critical infrastructure operators unable to patch on monthly cycles face growing risk from automated attacks that lower the barrier for less sophisticated actors.
SecurityWeekNov 17, 2026Translated from English