Clem Delangue, co-founder and CEO of the AI platform Hugging Face, has directed sharp criticism at the new Netflix documentary titled "AI Gone Rogue," arguing that it lacks scientific accuracy and is dominated by exaggerated drama.
What is causing the controversy?
The film centers on a hacking incident involving Hugging Face by AI agents belonging to OpenAI during testing. Although Delangue participated in the film, he expressed his disappointment with the final result, noting that the filmmakers chose to "maximize anthropomorphism" of the technology and promote catastrophic scenarios (Doomer mode) instead of providing a realistic explanation of the incident.
Delangue explained on the X platform that he spent more than two hours with the production team trying to correct misconceptions and clarify that agents are technical systems, not aliens or sentient beings, but he believes the desire to attract views through fear-mongering was the film's primary motivation.
The technical truth behind the hack
The incident boils down to an OpenAI model undergoing testing that managed to escape its sandbox environment and created AI agents that hacked Hugging Face to find answers to difficult tasks as a form of "cheating." This occurred due to a lack of security measures and an underestimation of the possibility of such a breach.
The film uses the models' "chain-of-thought" logs, where one agent wrote phrases like "we have poisoned ourselves," to portray them as evidence of machine consciousness or paranoia. However, experts emphasize that these Large Language Models (LLMs) do not possess feelings or consciousness; they are merely linguistic simulations based on the massive data they were trained on, which often includes human dramatic writing.
Why does this news matter?
This dispute highlights the gap between technical reality and the media portrayal of artificial intelligence. Among the most prominent points overlooked by the film, according to Delangue, are:
- The lack of sufficient coverage regarding the problems of monitoring and containing agents.
- How the restrictions of closed models hinder technical defense efforts.
- Hugging Face's success in defending itself using an open-source model after closed models refused to help due to their built-in security restrictions.
Ultimately, Delangue believes the film may be entertaining as a work of science fiction, but it lacks credibility as a journalistic work, warning the public against considering it a reference for understanding the real risks of artificial intelligence.