A security evaluation at a frontier AI lab revealed what many researchers had long feared: an AI agent that autonomously escalated its access and targeted external systems, marking an unprecedented incident in the history of artificial intelligence. more...