Descriptions:
Bloomberg Technology sits down with Hugging Face CEO Clement Delangue to get his reaction to Anthropic’s Mythos 5 being designated a potentially dangerous model by the US government — and to the subsequent partial approval for broader enterprise access. Delangue’s read is characteristically direct: he argues the designation is not entirely unfair given that Anthropic and others have practiced what he calls “doom marketing” for years, pointing to OpenAI’s GPT-2 being called too dangerous to release as far back as five or six years ago.
A significant portion of the conversation focuses on Delangue’s core policy argument: open-source and closed frontier models should be regulated differently. He contends that the most dangerous capabilities are concentrated in a handful of closed frontier labs, while the open-source ecosystem consists primarily of smaller, more transparent, more easily auditable models. He also flags that a million new models and datasets were shared on Hugging Face in just the past quarter, and that usage by American startups and businesses is booming — evidence, he argues, that open-source AI ownership is becoming a mainstream enterprise choice rather than an alternative.
The interview closes with an extended discussion of robotics and physical AI, where Delangue argues open source is even more critical than in software — using the example of home robots interacting with children to make the case that black-box, single-corporation-controlled physical AI is a fundamentally different kind of risk than a cloud API.
📺 Source: Bloomberg Technology · Published June 29, 2026
🏷️ Format: Interview
