China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university server
AI-assisted offensive security has crossed a notable threshold: attackers linked to China reportedly used autonomous agents to discover and exploit a previously unknown vulnerability, then compromise a university server. According to Amazon's CISO, the agents continued adapting for nearly a full day without human direction, suggesting AI was not merely generating code or recommending tactics, but participating in a sustained attack loop. For security teams, the practical concern is speed and persistence. Agents that can probe systems, adjust to failed attempts, and exploit a zero-day can compress work that once required hands-on expertise and continuous operator attention. The report does not identify the models, tooling, vulnerability, or affected university, so the degree of autonomy cannot yet be independently assessed. Still, builders deploying agents should treat tool access, network permissions, audit logs, and automated shutdown controls as core security boundaries, while defend
Read full article