White House Struggles to Decide on Chinese AI

The Trump administration is divided over how to respond to the rapid rise of leading Chinese AI models, according to sources familiar with the discussions. The dispute primarily pits parts of the White House, which advocate for stricter restrictions on Chinese AI that could soon rival the strongest US models, against the Department of Commerce, which deems such restrictions unenforceable. This issue has become a new point of tension for the administration in the AI space after Chinese lab Moonshot AI last week released the Kimi K3 model, which competes with top US models from Anthropic and OpenAI. In recent days, senior White House officials have discussed possible measures to prevent Chinese labs from developing their models by training on American models—a practice known as distillation. The White House is expected to take some presidential action after Anthropic last month accused Chinese tech company Alibaba of launching "the largest known distillation attack" against it. This action is not expected to take the form of an executive order. A White House spokesperson declined to comment on the matter. The White House takes distillation attacks particularly seriously. On Wednesday, it announced that Moonshot developed Kimi K3 by distilling Anthropic's Claude Fable 5 model, and Treasury Secretary Scott Bessent told Fox Business that covert distillation could lead to sanctions, calling it "intellectual property theft." While proposals are circulating inside the White House, a formal request has not yet been sent to the Department of Commerce, which has its own ideas, sources said. The Department of Commerce is becoming increasingly influential in the administration's internal divide over AI, partly because it controls export restrictions through the Bureau of Industry and Security. Lunik is mulling ways to encourage leading US labs to create their own open-weight models to counter China and spoke with executives from several AI labs last week. Lunik appears to take a middle-ground regulatory approach—he imposed export restrictions on Anthropic to keep them in check but acts more freely than Carncross, one source said. Earlier this month, the administration's biggest AI headache was Anthropic's Claude Mythos and Fable 5 models due to their ability to find vulnerabilities in government systems and critical infrastructure. But the political headaches from AI have only intensified with the emergence of so-called open-weight models from China—models that have little to no safeguards against hacking government systems and can be downloaded and used by anyone.