Google's Threat Intelligence Group reported that threat actors are using autonomous multi-agent AI frameworks to execute multi-stage cyber intrusions at machine speed. These systems leverage Large Language Models (LLMs) to automate the reverse engineering of software patches, allowing attackers to weaponize vulnerabilities without human intervention.
One specific framework, 'exploit_agent', was discovered on an exposed staging server containing stolen data and Russian-language comments. This tool is designed to automatically turn software patches into exploits, specifically targeting French media, entertainment, and education websites.