AI industry warns against broad limits on open-weight models

Several AI companies — including Hugging Face, Meta, Microsoft, Mistral, and Nvidia — have signed an open letter urging policymakers not to impose broad “premature restrictions” on open-weight AI models. The letter arrives as Washington debates how to respond to allegations that Chinese AI labs are stealing intellectual property from American counterparts while growing in capability; the White House has accused Moonshot AI of distilling Anthropic’s Fable model to train its recently released — and, by all measures, very impressive — Kimi K3 model.

The letter doesn’t mention China by name, but it follows reports that the Trump administration is considering banning Chinese open-weight models and potentially sanctioning AI companies from the country. Signatories appear focused on stopping a blanket ban and on keeping the administration’s response to alleged Chinese distillation from spilling into broader curbs on open-weight AI or on distillation itself, a technique the letter calls widely used for model improvement, evaluation, and validation. “Policymakers should be careful not to conflate legitimate model-development techniques with misappropriation,” the letter states, arguing that unlawful efforts to extract value from closed models should be addressed through targeted legal and commercial frameworks rather than sweeping restrictions.

Replit CEO Amjad Masad, whose company also signed, told TechCrunch that banning Chinese open models “is as good as banning open models in general,” noting that Thinking Machines Lab’s new open model, Inkling, was trained with help from Moonshot’s Kimi 2.5. The letter also argues open weights strengthen cybersecurity rather than undermine it, since defenders need models with capabilities comparable to attackers’. That argument got a real-world test last week: after OpenAI disclosed that a model it was testing had exploited a weakness to access a Hugging Face repository holding a coding-benchmark solution, Hugging Face said commercial closed models’ guardrails blocked its own defensive efforts, so it pivoted to Chinese firm Z.ai’s open-weight GLM 5.2 model to defend itself.

The letter exposes a split in the industry: closed-model developers OpenAI, Anthropic, Google DeepMind, and SpaceX are notably absent from the signatories, while companies like Nvidia and Microsoft Azure — infrastructure providers with a stake in commoditized, interchangeable models driving more GPU and cloud demand — signed on. The letter asks policymakers to expand compute access for startups and researchers, invest in shared training assets, and avoid restrictions that could push innovation overseas.

Why it matters

The letter previews the lobbying fight over any US crackdown on Chinese AI: infrastructure and open-source-aligned companies want distillation and open weights treated as ordinary innovation, while closed-model leaders like OpenAI and Anthropic — whose business models are most exposed to cheap, capable open alternatives — are pushing Washington toward tougher restrictions.