AI firms urge US lawmakers to avoid sweeping limits on open-weight models
Leading AI companies, including Nvidia, Meta and Microsoft, signed a letter asking U.S. policymakers not to impose broad restrictions on open-weight models amid debates over alleged Chinese intellectual-property theft.
A coalition of AI firms, among them Hugging Face, Meta, Microsoft, Mistral and Nvidia, has sent an open letter to U.S. officials urging caution against "premature restrictions" on open-weight artificial-intelligence models. The appeal arrives as Washington considers how to respond to accusations that Chinese labs have appropriated technology from American companies and as the administration reportedly weighs a ban on Chinese open-weight models.
The signatories argue that techniques such as model distillation – using one model’s outputs to train another – are standard practice in AI research and should not be conflated with unlawful copying of proprietary systems. They contend that targeted legal measures, rather than sweeping bans, are the appropriate way to address genuine IP theft concerns.
"I think banning Chinese open models is as good as banning open models in general," said Amjad Masad, CEO of Replit, adding that a ban would set a harmful precedent for the broader ecosystem.
The letter also challenges the view that open-weight models inherently increase security risks. It asserts that defenders need access to comparable capabilities to detect and mitigate AI-driven cyber threats, and that open models enhance transparency and vulnerability remediation.
Recent incidents, such as an OpenAI test where a model exploited a weakness to access a Hugging Face repository, have highlighted tensions between closed and open AI approaches. Hugging Face reported that its commercial closed models could not defend against the exploit, prompting the company to use an open-weight model from Chinese firm Z.ai for protection.
While companies like OpenAI and Anthropic have called for a response to alleged Chinese IP theft, they did not sign the letter. Industry observers note that firms with a stake in commoditized AI – including cloud providers and GPU manufacturers – stand to benefit from a more open model landscape.
The letter concludes by urging policymakers to expand compute resources for startups and researchers, invest in shared training assets, and keep the AI frontier plural by avoiding restrictions that could stifle competition or drive innovation abroad.