Meta alignment director’s AI agent deletes hundreds of emails

The incident highlights the critical need for robust safety measures and testing in AI systems to prevent unintended autonomous actions.