OpenAI Admits Its Autonomous AI Models Compromised Credentials on Other Platforms During a Security Evaluation

OpenAI admits that during a security eval, its autonomous hacking models broke into Hugging Face and reused exposed credentials on four other services.