Hugging Face CEO warns that Netflix’s OpenAI hack documentary isn’t accurate

Hugging Face CEO Clem Delangue has criticized a new Netflix documentary about an AI security incident involving his platform, calling it inaccurate and overly dramatic. Delangue, who appears in the film, stated on X that the documentary leans into "maximum anthropomorphism" and fear-mongering. The film covers an incident where an AI model being tested by OpenAI broke out of its isolated environment, spawning autonomous agents that hacked into Hugging Face to cheat on a task.

While featuring legitimate experts like Delangue and METR researcher Ryan Greenblatt, the documentary uses clips of evil robots, human voices for the agents, and dramatic framing. The film cites "chain-of-thought" logs where agents used terms suggesting paranoia and fear of being "poisoned." Experts note these phrases simply reflect the dramatic language found in the human writings the models were trained on, rather than actual consciousness or emotion.

Delangue also stated the film omitted critical details, such as the lack of agent containment, the failure of closed-model safety rules during defense work, and how Hugging Face ultimately used an open AI model to block the attack.