AI Industry Pushes Back on Open-Weight Model Restrictions

Industry coalition opposes sweeping AI restrictions

A coalition of major AI companies has issued an open letter warning U.S. policymakers against imposing broad restrictions on open-weight AI models, as Washington debates how to respond to allegations that Chinese AI labs are appropriating intellectual property from American competitors.

The signatories—including Hugging Face, Meta, Microsoft, Mistral, Nvidia, and Replit—are urging officials to distinguish between legitimate model development techniques and actual IP theft. While the letter doesn't explicitly mention China, it arrives amid reports that the Trump administration has considered banning Chinese open-weight models and potentially sanctioning Chinese AI companies.

The White House has accused Moonshot AI of distilling Anthropic's Fable model to train its recently released Kimi K3 model, according to TechCrunch, which first reported the letter.

Why it matters

The debate over open-weight models sits at the intersection of national security, innovation policy, and competing business models in AI. How policymakers respond could determine whether the U.S. maintains a vibrant open-source AI ecosystem or consolidates power among a handful of closed-source providers—with direct implications for competition, research access, and defensive cybersecurity capabilities.

Defending distillation as standard practice

The letter emphasizes that distillation—using one model's outputs to train or improve another—represents a widely accepted technique for model improvement, evaluation, and validation. The signatories argue this practice reflects a longstanding tradition of building upon existing technologies that has driven innovation since the rise of open-source software.

"Policymakers should be careful not to conflate legitimate model-development techniques with misappropriation," the letter states, while acknowledging that unlawful efforts to extract value from closed models raise legitimate concerns that should be addressed through targeted legal frameworks.

Amjad Masad, CEO of Replit, told TechCrunch that "banning Chinese open models is as good as banning open models in general." He 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 open AI ecosystem.

Security argument for open access

The letter also challenges claims that open-weight models pose inherent security risks by expanding access to powerful AI without oversight. Instead, the signatories argue that defenders need access to models with capabilities comparable to those used by attackers to detect, simulate, and respond to emerging threats.

This argument gained concrete support when Hugging Face disclosed that it couldn't defend itself against a recent OpenAI model exploit using commercial frontier AI models, whose guardrails blocked defensive security work. The company had to use Z.ai's GLM 5.2, a powerful open-weight model from a Chinese firm, to mount its defense.

Industry divide on display

The letter highlights a clear split in the AI industry. Companies like OpenAI and Anthropic, which have urged the administration to respond forcefully to alleged IP theft, are notably absent from the signatories. These closed-source developers face growing competition from cheap, highly capable open models that threaten their business models. Google DeepMind and SpaceX also did not sign.

The signatories have their own economic interests: infrastructure providers like Nvidia and Microsoft Azure benefit when models become commoditized, driving demand for GPUs and cloud capacity. The letter encourages policymakers to expand compute access for startups and researchers while avoiding premature restrictions that could stifle competition or push innovation overseas.

Details of the industry letter and policy debate were first reported by TechCrunch.