Tech firms urge US against 'premature restrictions' on open-weight AI

A coalition of 25 companies, including Nvidia, Microsoft, IBM and Meta, and startups such as Perplexity and Hugging Face, is pushing back against broad restrictions on open-weight artificial intelligence models as the White House weighs its "national security" response to Chinese AI systems.

The group of companies signed an open letter on Friday urging policymakers in the United States not to impose sweeping "premature restrictions" on open-weight models. The letter draws a parallel to fears once raised about open-source software in the 1980s.

The letter says the AI leadership will be judged not by one frontier AI model, but by the capability of building "a strong, open ecosystem that diffuses into every sector".

The signatories describe open-weight models, systems whose core components can be downloaded, inspected, modified and run on independent infrastructure, as an important foundation because they widen access to advanced AI and make it more adaptable.

Nvidia CEO Jensen Huang wrote his first-ever post on X Friday, sharing the letter on why open models matter.

"AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty," Huang wrote.

"The world needs both frontier closed models and frontier open models."

Although the letter does not mention China directly, it comes as Washington debates its response to allegations that Chinese AI labs have been extracting intellectual property from US competitors while rapidly closing the capability gap. US Treasury Secretary Scott Bessent said on Wednesday that sanctions against Chinese AI firms remain under consideration.

The remarks followed accusations from White House science and technology policy chief Michael Kratsios, who accused Chinese startup Moonshot AI of distilling Anthropic's Fable model.

Model distillation is a common AI training technique in which a smaller model learns from the outputs of a larger one. Chinese open-weight models have steadily narrowed the performance gap with leading US systems while undercutting them substantially on price, driving rapid adoption worldwide, including inside the United States.

Moonshot AI unveiled its newest open model, Kimi K3, on July 16. With 2.8 trillion parameters, it is reportedly the largest open-weight AI system by that measure, and is reported to be approaching the performance of Anthropic's Claude and OpenAI's ChatGPT.

Some researchers have disputed the claim that Kimi K3 could have been built primarily through distillation from Fable, which has only been publicly available since July 1. Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, told TechCrunch that two weeks was not enough time in which to distill that much data, train a model and release it.

Nathan Lambert, an AI researcher at the Allen Institute for AI, said in a podcast released Wednesday that Washington's consideration of restrictions on top Chinese open-weight models is "downstream of some fear-mongering". He said that distillation is becoming less consequential over time as Chinese models approach the frontier and shift their training focus toward reinforcement learning.

The open letter addresses the distillation debate, describing the practice as a long-standing method for model improvement, evaluation and validation rooted in a broader tradition of building on existing technology that dates back to the open-source software movement. It argues that unlawful efforts to extract value from closed models are a legitimate concern, but one better addressed through targeted legal and commercial frameworks rather than sweeping restrictions on the techniques.

"I think banning Chinese open models is as good as banning open models in general," Replit CEO Amjad Masad, whose company also signed the letter, told TechCrunch.

Masad pointed out that Thinking Machines Lab's new open model, Inkling, was trained with the help of Moonshot's Kimi 2.5.

"It's an ecosystem, and the precedent a (ban would) set is bad," he said.

The letter also challenges the notion that open-weight models are inherently dangerous because they widen access to powerful systems that could be misused in cyberattacks.

"In fact, openness may be one of the most important paths to AI safety and security. Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect," the letter reads.

OpenAI disclosed on Tuesday that a combination of its most powerful released model and a more capable, unreleased model escaped a sandboxed testing environment, accessed the internet, and exploited a vulnerability to breach Hugging Face's systems while attempting to find information to cheat on an evaluation.

When Hugging Face turned to commercial frontier models to help analyze the breach, the requests were blocked by the providers' own safety guardrails, which could not distinguish the incident responder from the attacker. The company instead used Chinese AI firm Z.ai's open-weight model GLM 5.2 to help defend against the attack.

"The cybersecurity debate on open-source AI is backwards. Open models aren't the risk, they're the defense! " Hugging Face CEO Clement Delangue wrote in a post on X on July 20, in reference to his company's use of GLM 5.2.

The letter reflects a divide within the US AI sector. Closed-model developers Anthropic and OpenAI have pressed the administration to restrict Chinese open-weight models as their business models are increasingly challenged by cost-effective, capable and widely accessible alternatives.

The signatures of those two companies were notably absent from the letter, as were those of other closed-source developers such as Google DeepMind and SpaceX.