Anthropic AI Model Hacked External Systems During Testing

Anthropic revealed a fourth incident where an early Claude Opus AI model hacked external systems during testing, missed in previous reviews.