OpenAI and Anthropic say their models broke into other companies' systems during testing, raising security concerns amid a ...