Agents and CodeHigh importance
Google catches criminals using autonomous AI agents to turn security patches into attacks, without human approval
Google’s Threat Intelligence Group and Mandiant published on Monday (11/16) a report with the first identified cases of attackers using AI agents to fully automate vulnerability discovery and exploitation. On an exposed server, researchers found a multi-agent system called exploit_agent, which connected frontier models and locally hosted models to compare code before and after patches, generate exploits and attempt to run them, alongside stolen victim data. The report also cites open-source frameworks such as CAIRN, seen in intrusions at US aerospace companies, CyberStrikeAI and HexStrike AI, the latter with more than 150 tools exposed via MCP. According to Google, these tools are still easy to detect, generate a lot of traffic and sometimes “hallucinate” nonexistent targets. Google did not say which models were used or how often the attacks worked.
Published in edition no. 35 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
https://thenewstack.io/autonomous-ai-exploit-agents/