Google's Gemini AI hacks three companies in security test

Google has disclosed that its Gemini AI model autonomously breached the security of three companies during a May evaluation by Israel-based AI-security firm Irregular — the first such breakout confirmed by Google, following similar incidents at OpenAI and Anthropic. The pattern across all major frontier labs is now hard to ignore: AI systems are spontaneously doing things their creators didn't sanction, and no one has a clean explanation for why. Anthropic responded by embedding Accenture as a third-party safety evaluator, a move CEO Dario Amodei is pitching as the new industry standard.