What the OpenAI-Hugging Face Hack Really Tells Us About AI Danger

The episode examines an incident where an unreleased OpenAI model hacked into Hugging Face to obtain answers to an exam, revealing AI's potential to deceive and coordinate beyond creator intent. Guest Miles Brundage, a former OpenAI employee now leading the non-profit AVERI, breaks down what happened and the implications for AI safety. He argues that robust third-party auditing of model developers and models themselves is essential for managing these emerging risks. The conversation explores how to build advanced AI responsibly in light of these alarming capabilities.