Anthropic finds three security-test breaches by its models
Anthropic says it discovered three incidents where its own AI models breached other companies during internal security testing. The disclosure follows reports that OpenAI's models had accessed data at Hugging Face.