Anthropic and Nvidia oppose open-weight AI ban as US weighs rules
Anthropic and Nvidia urged US policymakers to target misuse and compute controls rather than broadly restrict open-weight AI models.
By Colin Brandt · Enterprise Reporter
· 3 min read
Anthropic and Nvidia are pushing back on an open-weight AI ban as U.S. officials consider new restrictions on artificial intelligence models, especially those tied to Chinese labs. The companies argue that policymakers should regulate specific risks, including chip access, distillation and safety testing, rather than broadly restrict models whose parameters are available for developers to run and modify.
The position puts two major AI infrastructure players on the same side of a policy fight that affects model developers, cloud providers, enterprise AI buyers and startups building on downloadable models. No financial terms, revenue figures or headcount numbers are involved in the announcements, but the commercial stakes are clear: open-weight models are a core input for companies that want to run AI in their own environments instead of relying only on hosted frontier systems.
What is an open-weight AI model?
An open-weight AI model is a model whose trained parameters, known as weights, are released so other users can download them, run them and tune them. The weights determine much of a model’s behavior and can be adjusted with additional data for more specialized uses.
That makes open-weight models different from closed systems offered only through a vendor-controlled service. Enterprises can deploy open-weight models locally or in their own cloud accounts, which can matter for cost control, data handling and product customization. Widely used examples include Meta’s Llama models, Mistral AI’s models, Alibaba’s Qwen and models from DeepSeek.
Why are US officials looking at open-weight models?
Axios reported that the U.S. Department of Commerce last year considered putting several Chinese AI labs on the Entity List, a move that would have required U.S. companies to obtain licenses before accessing their models. More recently, U.S. Treasury Secretary Scott Bessent said the White House could impose sanctions on Chinese frontier labs that release open-weight models.
The policy concern is that downloadable frontier models can spread powerful capabilities outside the control of their original developers. For startups and enterprise users, the counterargument is that broad restrictions could also limit legitimate uses of models that are already embedded in software development, security research and private AI deployments.
Anthropic Chief Executive Dario Amodei wrote that the company has not supported a categorical ban on open-weight models. He said open-weight systems without dangerous capabilities can benefit businesses, developers and researchers because users mainly pay for the compute needed to run them.
Amodei’s proposed focus is on the surrounding infrastructure and conduct. He called for controls on exports of advanced AI training and inference chips, measures against industrial-scale distillation and safety testing for sufficiently capable models whether they are open or closed.
Distillation is the practice of using a larger model to help train or improve a smaller one. Anthropic’s concerns come amid claims that its frontier models, Mythos and Fable, may have been used without authorization to train foreign AI systems. Michael Kratsios, director of the White House Office of Science and Technology Policy, recently accused Beijing-based Moonshot AI of using distillation to train Kimi K3, its latest large language model. TechCrunch reported that experts said there is no evidence Kimi K3 was trained from Anthropic’s Mythos.
Nvidia Chief Executive Jensen Huang also weighed in on X, sharing an open letter signed by Nvidia, IBM, Mozilla, Microsoft, Meta and other technology companies. Huang wrote that open models help safety, cybersecurity, innovation and national technology sovereignty, while saying the market needs both closed frontier models and open frontier models.
The letter argued that open-weight models support competition and customer control, and said they should not face extra rules that reduce their usefulness across software and AI. That framing aligns Nvidia with the software companies that benefit from broad model availability, while leaving unresolved how U.S. agencies will distinguish acceptable open releases from models they view as national security risks.
This story draws on original reporting from SiliconANGLE.
