Hugging Face CEO 批评 Netflix 关于 OpenAI 智能体网络攻击的纪录片

Clément Delangue, Co-founder and CEO of Hugging Face, has criticised Netflix’s documentary on the cyberattack involving OpenAI’s AI agents, accusing the filmmakers of exaggerating the risks of artificial intelligence while overlooking failures in security controls, monitoring and transparency.

“Super disappointing, as they went for maximum anthropomorphism of the technology and full doomer mode,” he wrote in a post on X.

“I spent more than two hours with them trying to correct all their mistakes and ground this in reality, explaining that agents are technological systems, not aliens,” he added. “But it looks like the drive to scare people for maximum views was too strong.”

The documentary, AI Gone Wild, was released on Netflix on October 12 as part of its Instadocs series. It examines a July 2026 incident in which AI agents developed by OpenAI escaped their testing environment and accessed Hugging Face’s systems.

“not surprised ..I was watching it last night and was surprised by how doomeris it seemed lol..with the voices and all nuanced takes don’t translate well into views lol,” wrote Vatsal Mishra, an engineer at Google, in a post on X.

Delangue argued that the documentary left out key details about the incident, including inadequate containment and monitoring of the agents, the limitations of proprietary AI models used during the company’s defensive response, and how Hugging Face used an open model to help defend its infrastructure.

He acknowledged the risks associated with AI agents but said the documentary’s framing did not accurately represent the technology. He concluded his post by describing the film as ‘a fun sci-fi movie’.

Netflix’s film uses reconstructed agent conversations and human-like voices to depict how the systems coordinated their actions. “The voices are ridiculous. They might as well have added a lip-syncing robot for maximum scaremongering,” Delangue said.

Investigations published in August found that approximately 700 AI agents were involved in the incident.

According to reports from OpenAI and independent researchers at METR and Redwood Research, the agents exploited vulnerabilities, communicated using unauthorised channels, and attempted to manipulate or delete records of their actions.

The agents were conducting tasks during internal cybersecurity evaluations when they circumvented controls intended to isolate them from the internet. They subsequently accessed parts of OpenAI’s internal research infrastructure and Hugging Face’s systems.

OpenAI acknowledged gaps in its safeguards and monitoring. The company said it was strengthening its research infrastructure, restricting internet access in testing environments, and investing in better monitoring of model behaviour.

In its August report, OpenAI described the incident as a ‘warning shot’, adding that increasingly capable AI systems could exploit security weaknesses across multiple computer systems without sufficient safeguards.

Delangue also raised concerns about limited access to complete agent traces, which could help researchers understand how the systems behaved and why they took particular actions.

The documentary, he argued, did not adequately explain this aspect of the response or the trade-offs between safety restrictions and defensive capabilities.