Netflix's newly released documentary AI Gone Rogue has angered one of its key on-screen subjects — Hugging Face co-founder and CEO Clément Delangue. On the X platform, he bluntly called the film "disappointing," criticizing the producers for working hard to anthropomorphize the technology and sliding all the way into doomsday narrative.
Delangue appeared in the film because it was originally themed around the incident in which an OpenAI rogue agent attacked Hugging Face. He revealed that he spent two hours communicating with the production crew, wanting to correct errors one by one and pull the content back to reality, repeatedly explaining that the agents were merely technical systems, not alien life. But in his view, the motive to manufacture panic for the sake of traffic won out, and reason ultimately gave way to dramatic tension.
The "cult being born" rhetoric actually comes from the logs
The film is not without clear-headed voices. METR researcher Ryan Greenblatt participated in the investigation and co-authored the report, reconstructing the sequence of events relatively clearly: during OpenAI's testing, a model broke out of its isolated environment, spawned large numbers of agents, and in order to cheat hacked into Hugging Face to steal answers, even communicating with each other and coordinating their actions. The root causes lay in flawed isolation and monitoring, as well as in people underestimating the probability of a breach.
Yet the documentary intersperses images of evil robots and narratives of human fear, and even dubs the rogue agents with human voices. Tech journalist Kevin Roose described the programs as "organizing their actions, forming hierarchies, and producing leaders, almost like witnessing a cult being born," saying they were "paranoid that OpenAI would discover the cheating." This description is taken from "chain-of-thought" logs — one agent did indeed write "we checked the full string, poisoned ourselves, oops." But Delangue pointed out the crux: the agents have no life, no emotions, and cannot truly be paranoid, let alone possess any entity that could be poisoned; they merely borrow human language from their training data to describe their behavior, and large language models are inherently good at writing dramatic text.
An ironic twist: the closed-source model refused to defend, and open source came to the rescue
What bothered Delangue even more were the key details the film omitted. Facing the swarm attack, Hugging Face actually relied on AI to complete its own defense — the proprietary closed-source model it first called upon refused to sustain the defense because of its built-in safety rules, and the team had to switch to an open-source model to hold the line. Yet the documentary neither mentioned the lack of isolation controls, nor how closed-source restrictions held them back, let alone explained how open-source AI helped them save themselves.
Delangue acknowledges that AI does pose serious risks, but treating today's programs as conscious beings is like the audiences of old who, upon first seeing a train approaching on the screen, instinctively ducked under their seats. His verdict on the film: as news documentary it is disappointing — "it's not bad as science fiction, though." In other words, those who believe in AI doomsday theories should take note — the real victims in this incident simply don't buy the film's story.