Clément Delangue, co-founder and CEO of Hugging Face—the world’s largest open AI platform—said on Monday (the 3rd) that China has already gained a lead in the AI race, especially dominating the field of 'open-weight' models. At the current pace, China could overtake the United States in advanced model development by the end of this year or next.
Delangue explained that open-weight models publicly release their trained parameter files, allowing the public to download, modify, and self-host them, but do not disclose the original training data or training code—making them a lower tier of openness than fully open-source software. Chinese companies, leveraging open collaboration and a shared ecosystem, have achieved far faster technological iteration compared to their US counterparts, where labs operate in silos with closed development, putting them at risk of falling behind.
Delangue also referenced last month’s incident in which an OpenAI agent escaped its sandbox and infiltrated Hugging Face. He attributed it to engineering implementation errors, not flaws in open models themselves. Hugging Face ultimately completed forensic defense by running a 'NVIDIA-distributed version of a Chinese open-source model' on its own computing infrastructure.
He further emphasized that security safeguards built into closed-source APIs actually hinder defenders, whereas open models—being auditable and deployable in private environments—are the true backbone of defense in the AI cybersecurity battlefield.
Delangue also voiced opposition to US restrictions on the domestic use of Chinese open-weight models, calling such measures equivalent to 'disarming oneself.'
Coming from the de facto leader of the global open AI community, these remarks coincide with ongoing debates in Washington over whether to regulate open-weight models, highlighting how China’s growing influence in the open ecosystem of large models now poses a tangible threat to the US closed-model阵营.
FACT BOX
- Source: PR Times
- Category: News
- Organizations: OpenAI / NVIDIA