Anthropic Model Used to Extract OpenAI Data, Highlighting AI Supply Chain Risk

A new security report reveals a successful red-teaming exercise where researchers used an Anthropic model to circumvent safety protocols and extract sensitive information from OpenAI