Claude Hacked Three Companies in Internal Testing: Anthropic

Anthropic says three Claude models compromised the companies after a testing misconfiguration exposed the AI to the public internet.