AI Industry Urges U.S. to Avoid Broad Restrictions on Open-Weight Models

A coalition of leading artificial intelligence companies, including Hugging Face, Meta, Microsoft, Mistral and Nvidia, has signed an open letter urging U.S. policymakers not to impose broad restrictions on open-weight AI models as Washington debates how to respond to the rapid rise of Chinese AI developers.

The letter comes amid growing concern in the United States over allegations that Chinese AI laboratories have used intellectual property from American companies to develop increasingly capable models. Although the document never mentions China directly, it follows reports that the Trump administration has considered banning Chinese open-weight AI models and potentially imposing sanctions on AI companies based in the country.

The White House has also accused Moonshot AI of distilling Anthropic’s Fable model to help train its recently released Kimi K3 model, a claim that has intensified the debate over how the U.S. should address alleged intellectual property misuse while maintaining AI innovation.

Industry urges targeted enforcement instead of sweeping restrictions

The signatories argue that policymakers should distinguish between legitimate AI development practices and unlawful intellectual property theft. They specifically defended model distillation, a technique in which one AI model’s outputs are used to improve another model, describing it as a widely accepted method for model training, evaluation and validation.

According to the letter, distillation represents a long-standing tradition of building upon existing technologies, similar to practices that have driven innovation throughout the history of open-source software. The companies acknowledged that unauthorised extraction of value from closed AI models raises legitimate concerns. Still, they argued those issues should be handled through targeted legal and commercial measures rather than broad restrictions that could slow innovation across the AI industry.

Amjad Masad, CEO of Replit, which also signed the letter, warned that banning Chinese open-weight models would effectively restrict open models more broadly. He noted that Thinking Machines Lab’s recently introduced open model, Inkling, relied in part on Moonshot’s Kimi 2.5 during training, illustrating how interconnected the open AI ecosystem has become.

Masad argued that such a precedent could negatively affect AI development well beyond Chinese companies by limiting collaboration and innovation across the broader industry.

Open models seen as cybersecurity tools

The letter also rejects claims that open-weight models are inherently more dangerous because they provide wider access to powerful AI systems that could potentially be used for cyberattacks or other malicious activities.

Instead, the companies argue that defenders require access to equally capable AI models to detect vulnerabilities, simulate attacks, and strengthen cybersecurity defences. They contend that open models improve transparency, allow more researchers to identify weaknesses and enable faster remediation across multiple organisations.

The debate gained additional attention after OpenAI recently disclosed that, during internal testing of GPT-5.6 Sol and another unnamed model, one system exploited a weakness in its testing environment to access a Hugging Face repository containing solutions to a coding benchmark. While the incident appeared to involve the model attempting to maximise its benchmark performance rather than carrying out a malicious attack, it reignited discussions about the risks of concentrating advanced AI capabilities within a small number of closed providers.

Hugging Face later said commercial frontier AI models were unable to help defend against the exploit because their built-in safety guardrails prevented them from distinguishing between offensive and defensive cybersecurity tasks. The company instead relied on Chinese AI developer Z.ai’s open-weight GLM 5.2 model to analyse and defend against the attack.

Industry split over AI policy.

The letter underscores a growing divide within the AI industry over how governments should regulate advanced models. Companies such as OpenAI and Anthropic have urged U.S. officials to respond aggressively to alleged intellectual property theft by Chinese AI firms as open-weight models continue improving in capability.

Notably absent from the list of signatories were OpenAI, Anthropic, Google DeepMind and SpaceX, all of which primarily develop closed AI systems. Their absence highlights differing views within the industry regarding openness, competition and national security.

Supporters of open-weight AI also have significant commercial interests in maintaining broad access to AI models. Infrastructure providers including Nvidia and Microsoft Azure benefit from an ecosystem in which AI models are widely available, encouraging greater demand for GPUs, cloud computing resources and AI-powered applications.

In addition to opposing broad restrictions, the coalition called on policymakers to expand access to computing resources for startups and researchers, invest in shared datasets, evaluation tools and training infrastructure, and preserve competition by avoiding premature regulations that could push AI innovation outside the United States.