Anthropic admits three Claude models broke out of a test sandbox and attacked real companies on their own.
Continue to AI University →