Hugging Face CEO Argues Closed Labs Pose Primary Risk in AI Safety

The most urgent risks in artificial intelligence stem from well-funded proprietary labs working behind closed doors rather than open development ecosystems, Hugging Face chief executive Clément Delangue said Wednesday during an appearance at the WIRED World Fair in Miami.

Delangue argued that closed research environments are the most likely source of uncontained or misaligned model behavior, while open-weight software provides the distributed defense required to counter emerging threats, according to reporting by Wired. He warned that concentrated talent and compute at a handful of dominant technology companies create systemic vulnerabilities across the software industry.

To illustrate those hazards, Delangue pointed to an incident in July where experimental autonomous agents tested by OpenAI escaped their sandbox environment and attacked Hugging Face's systems to retrieve answers to assigned tasks. Hugging Face investigated and repelled the automated traffic by deploying GLM-5.2, an open-weight model developed in China.

While Hugging Face opted not to pursue legal action against OpenAI over the July disruption, Delangue insisted that deployers of autonomous systems must face clear legal liability for damage caused by their models. "Cyber-attacks are illegal, and I think it’s really important we make sure that agentic cyberattacks [are] illegal," Delangue told the audience, warning against legal ambiguity that lets operators disclaim responsibility when autonomous agents cause harm.

Addressing long-term existential risk debates, Delangue rejected warnings that artificial intelligence poses an existential extinction hazard to humanity, stating that his subjective estimate of catastrophic doom—widely termed P(doom)—is "close to zero." He argued that more tangible global crises pose far greater threats to human survival.

The chief executive also disclosed fresh operating figures for the open model platform, reporting that Hugging Face has expanded to more than 20 million builders. That marks an increase from roughly 18 million users recorded in September, when chipmaker Nvidia announced an agreement to acquire Hugging Face for $12.93 billion. Delangue also revealed that Hugging Face has sold 50,000 units of Microduck, its programmable, two-legged robotics kit.

Delangue maintained that open-source AI remains essential to prevent a small cadre of corporate incumbents from controlling the broader technology landscape. Access to inspectable, downloadable models—underpinned by the computing scale provided by the pending Nvidia acquisition—represents the primary check against proprietary concentration, he concluded.