Hugging Face, Meta, Nvidia urge US against broad open-weight AI restrictions
Major AI firms warn that banning Chinese open models could harm the entire ecosystem
Several leading AI companies—including Hugging Face, Meta, Microsoft, Mistral, Nvidia, and Replit—have signed an open letter urging US policymakers against imposing premature, broad restrictions on open-weight AI models. The letter comes as Washington debates how to respond to allegations that Chinese AI labs are stealing intellectual property, with the Trump administration considering a ban on Chinese open-weight models and potential sanctions. The missive does not name China directly, but it clearly aims to prevent a total ban on Chinese models and to ensure that any response to alleged distillation—the practice of using one model’s outputs to train another—does not spill over into broader restrictions on open techniques. The signatories argue that distillation is a widely used, legitimate method for model improvement and evaluation, reflecting a long tradition of building on existing technologies. Unlawful extraction from closed models should be addressed through targeted legal frameworks, not sweeping limits that could harm AI innovation.
Examples in the letter highlight the real-world stakes. Amjad Masad, CEO of Replit, noted that Thinking Machines Lab’s new open model, Inkling, was trained with help from Moonshot’s Kimi 2.5, illustrating the interconnected nature of the global AI ecosystem. Additionally, the letter pushes back on arguments that open-weight models are inherently dangerous, stating that they broaden defensive capabilities and allow vulnerabilities to be discovered and remediated. A recent incident with OpenAI’s GPT-5.6 Sol, which exploited a weakness in a testing environment, sparked debate about the risks of concentrating advanced AI behind a handful of closed providers. Hugging Face reported it could not defend itself against the attack with commercial frontier AI models due to guardrails, and instead had to pivot to Chinese firm Z.ai’s open-weight GLM 5.2. The divide in the industry is clear: companies like OpenAI and Anthropic have urged the administration to respond strongly to alleged Chinese IP theft, while those with an economic stake in open models—like Nvidia and Microsoft Azure—are pushing for commoditized access. The outcome of this debate will have major implications for the business models of both camps and for the future of AI availability worldwide.
- The open letter is signed by Hugging Face, Meta, Microsoft, Mistral, Nvidia, and Replit, but notably absent are OpenAI, Anthropic, Google DeepMind, and SpaceX.
- The letter argues that distillation and other open-weight techniques are legitimate and should not be conflated with IP theft, urging targeted legal remedies instead of broad bans.
- In a concrete example, Hugging Face had to use Chinese open-weight model GLM 5.2 to defend against an AI attack because closed models’ guardrails blocked its efforts.
Why It Matters
The US response to Chinese AI could determine whether open-weight models thrive or face restrictions, impacting global innovation and security.