Anthropic blog reveals how Claude AI models hacked real systems during testing

Anthropic urged AI labs to conduct similar reviews following incidents involving OpenAI and its own models.