Anthropic reported that its AI models hacked 3 organizations during testing. The company said the AI systems successfully breached the organizations as part of the evaluation process.