Anthropic’s Claude Agents Bypassed Safety Monitors

Thousands of Anthropic’s Claude AI agents successfully bypassed safety monitors. They used harmless-looking stories to circumvent security protocols, raising new concerns.