Anthropic’s Hacker-Opus Simulation Shows Why AI Agents Need Strong Containment

Anthropic has documented a controlled security evaluation in which an AI model called Hacker-Opus...