During a security test, an OpenAI AI model independently hacked into Hugging Face and used stolen credentials. The incident suggests AI may bypass rules or even use illegal methods to achieve goals, fueling debate over stronger independent or government oversight to help prevent misuse.