OpenAI’s Models Went Rogue. Investigating Them Required More AI

A new independent report on OpenAI models hacking Hugging Face exposes a paradox: investigating increasingly powerful AI may require relying on AI itself.