Google confirms case where AI automated everything from patch analysis to attack execution
Original headline구글, AI가 패치 분석부터 공격 실행까지 자동화한 사례 확인
Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.
Google Threat Intelligence Group and Mandiant reported on the 16th that they had confirmed the first real-world case of an AI system autonomously finding and exploiting a vulnerability, with an "exploit_agent" installed on an internet-exposed attack-preparation server. The system chains frontier models with multiple local large language models and continuously compares source code of a popular web application before and after patches to spot vulnerabilities in the changed code, then writes and executes exploit code without human intervention. The finding matters because it marks a shift from AI-assisted security work to end-to-end automated discovery and exploitation in live threat activity. What to watch next is whether such agent-driven exploitation spreads across more exposed servers and how quickly defenders can patch before AI agents weaponize the changes.
Muhabbit's AI wrote this summary in its own words from TokenPost's report and translated it from Korean. It can get details wrong, so check the source.
Named in this story
Google (GOOGL), Mandiant, Google Threat Intelligence Group, exploit_agent
A linked name opens that asset's sentiment page.