Hugging Face CEO Warns China is Winning AI Race Through Open-Source Collaboration
Hugging Face CEO Clément Delangue stated that China is winning the AI race due to its collaborative open-source ecosystem, while US labs remain closed and fragmented. He highlighted a recent incident where Hugging Face used Zhipu AI's open-source GLM 5.2 model to defend against an attack from an unreleased, closed-source OpenAI model. This highlights the growing geopolitical debate over open-weight AI models and suggests that open-source collaboration may accelerate technology development faster than proprietary, closed-source approaches. It also underscores the critical role of open models in cybersecurity defense against rogue or unreleased AI systems. Delangue argued that closed-source models with API guardrails cannot be effectively used for cybersecurity defense, making open-weight models essential for protecting systems. Meanwhile, US tech giants like Nvidia, Meta, and Microsoft have actively lobbied the US government against restricting open-weight AI models.
## BACKGROUND
Open-weight AI models allow developers to access and modify the underlying neural network weights, enabling local deployment and customization. Zhipu AI's GLM 5.2 is a prominent open-weight Mixture-of-Experts (MoE) model released in mid-2026, designed for complex coding tasks and long-context processing. AI guardrails are safety constraints embedded in AI systems to prevent harmful outputs, but they can restrict how developers interact with the model during security incidents.