A coalition of more than 25 technology companies led by Nvidia, Meta, and Microsoft has urged U.S. policymakers not to restrict Chinese open-weight AI models, warning that limits designed to curb competitors would hurt American innovation. Open-weight models release their internal parameters publicly, allowing anyone to run and modify them, and the companies' joint letter argues premature restrictions would concentrate advanced AI behind a few closed systems.
Moonshot AI raised the stakes earlier this month by releasing Kimi K3, a Chinese model that outperformed several American competitors on industry benchmarks. White House advisor Michael Kratsios alleged Moonshot built it by distilling Anthropic's models, using their outputs to train a cheaper system.
The companies defended distillation as standard practice in AI development. The letter drew a line between legitimate use and misappropriation, calling
for targeted legal frameworks rather than broad restrictions. It also argued that concentrating advanced AI behind a few closed systems creates failures outsiders can't detect.
Neither OpenAI nor Anthropic signed the letter. OpenAI president Greg Brockman said he had not been involved in any White House talks about restricting Chinese models. |