Google reports that, for the first time, it has seen attackers use an AI framework to find and exploit software flaws with no human involved. The findings reference offensive AI tools including HexStrike AI, PentAGI, CyberStrikeAI, Agent Zero, and ZeroHollow, along with state-backed groups Google tracks as Calanque Ion (formerly APT42), Leaky Castle and Midnight Neptune. Here is what IT and security teams should know about these findings.