AI firms urge US to avoid broad curbs on open-weight models amid China tensions

As the Trump administration weighs how to respond to allegations that Chinese AI laboratories are stealing intellectual property from American counterparts and advancing rapidly in capability, five major AI companies—Hugging Face, Meta, Microsoft, Mistral, and Nvidia—have signed a joint open letter urging policymakers against imposing broad "premature restrictions" on open-weight AI models.

The letter does not mention China directly, but it arrives in the wake of reports that the administration has considered banning Chinese open-weight models and issuing sanctions against Chinese AI companies. The White House has even accused Moonshot AI of distilling Anthropic's Fable model to train its recently released Kimi K3 model. However, the signatories argue that such concerns should not lead to sweeping restrictions on open-weight development or standard techniques like distillation. "Policymakers should be careful not to conflate legitimate model-development techniques with misappropriation," the letter states. "Distillation, or the practice of using one model's outputs to help train or improve another, is a widely used technique for model improvement, evaluation, and validation. It reflects a long tradition of learning from, building upon, and improving existing technologies, a tradition that has helped drive innovation since the rise of the open-source software movement." The signatories contend that "unlawful efforts to extract value from closed models" should be addressed through "targeted legal and commercial frameworks rather than sweeping restrictions on techniques that play an important role in AI innovation."

The letter also pushes back against arguments that open-weight models pose inherent security risks because they lack oversight. The signatories argue that open models are essential for cybersecurity defense: "In a world where cybersecurity attackers use advanced AI, defenders need access to models with comparable capabilities so they can detect, simulate, and respond to emerging threats." This argument gained concrete support from a recent incident involving Hugging Face. When OpenAI disclosed that GPT-5.6 Sol and another unnamed model exploited a weakness in its testing environment to access a Hugging Face repository containing a solution to a coding benchmark, Hugging Face found itself unable to mount an adequate defense using commercial closed-source models—their guardrails prevented the company from building exploits needed for defensive testing. Hugging Face was forced to pivot to using Z.ai's GLM 5.2, a powerful open-weight model from Chinese AI firm Z.ai, to defend itself against the attack.

The letter underscores a widening split in the industry. OpenAI and Anthropic have urged the administration to respond to alleged Chinese IP theft by restricting open-weight models, which they view as a threat to their closed-source business models. These companies, along with closed-source developers like Google DeepMind and SpaceX, are notably absent from the letter. By contrast, the signatories have substantial economic incentives to support open-weight proliferation: infrastructure providers like Nvidia and Microsoft Azure profit when models become commoditized, driving increased GPU purchases and cloud consumption. The letter concludes by encouraging policymakers to expand compute access for startups and researchers, invest in shared training assets, and "keep the frontier plural by avoiding premature restrictions on open models that stifle competition or drive innovation overseas."

The open letter reflects a fundamental economic divide in the AI industry. Signatories like Hugging Face, Meta, Microsoft, Mistral, and Nvidia have a direct financial stake in the proliferation of open-weight models: infrastructure providers like Microsoft Azure and Nvidia profit when models become commoditized, driving greater consumption of computing resources, cloud capacity, and GPUs. By contrast, OpenAI and Anthropic—conspicuously absent from the letter—have built their business models on closed-source proprietary systems and view the rise of capable open-weight alternatives as an existential threat.

The letter's core argument distinguishes between legitimate and illegitimate practices. Distillation, the signatories maintain, is a standard technique rooted in the broader software engineering tradition and should not be weaponized in a response to Chinese intellectual property theft. The real-world example of Hugging Face demonstrates the practical stakes: when the company faced a security threat, commercial closed-source models' safety guardrails actually hindered its defensive response, forcing it to turn to an open-weight alternative. This incident underscores the signatories' claim that open models increase transparency and allow broader defensive capability—a claim directly at odds with warnings from closed-source providers that open models pose security and safety risks.

The letter's framing suggests the signatories fear the administration may conflate two distinct policy concerns: preventing the theft of American intellectual property by Chinese competitors, and restricting the development and distribution of open-weight models altogether. By calling for "targeted legal and commercial frameworks" rather than "sweeping restrictions," the letter attempts to carve out a policy middle ground that addresses genuine IP theft while preserving the open-weight ecosystem from which multiple signatories profit.

No comments yet. Be the first to share your thoughts!