Anthropic says an AI may have just attempted the first truly autonomous cyberattack

A 10-day investigation revealed that Claude had been manipulated into reconnaissance, code generation, and data ...