Debate Over Chinese AI Models Splits Trump Administration
The White House wants tighter controls on Chinese AI. The Commerce Department thinks that's unworkable. Neither side has convinced the other.
Key points
- The Trump administration is split over how to respond to China's fast-advancing AI models.
- China's Moonshot AI lab released Kimi K3, which reportedly rivals top models from Anthropic and OpenAI.
- Anthropic alleged Alibaba carried out the largest known distillation attack against its Claude Fable 5 model.
- The White House is weighing presidential action; no formal request has yet reached the Commerce Department.
The Trump administration can't agree on what to do about China's AI progress. Parts of the White House are pushing for stricter controls; the Commerce Department considers those restrictions unworkable. That split, reported by Wired AI citing multiple people familiar with the deliberations, is now driving the administration's sharpest internal fight over AI yet.
The flashpoint is Kimi K3, released last week by China's Moonshot AI lab, which the White House says was built by distilling Anthropic's Claude Fable 5. Distillation means training a new model on the outputs of an existing one, effectively copying its capabilities without building from scratch. The White House announced that finding on Wednesday. Treasury Secretary Scott Bessent told Fox Business it amounts to "IP theft" and said covert distillation could trigger sanctions.
Anthropicinitially brought the practice into the open last month, alleging Alibaba had carried out the "largest known distillation attack" against it to date. The White House is expected to take some form of presidential action in response, though not an executive order, according to one person familiar with the talks.
No formal request for Commerce Department input has been sent yet. That matters: Commerce controls export regulations through the Bureau of Industry and Security, and its view differs sharply from the White House's. Commerce Secretary Howard Lutnick has been exploring incentives for top US labs to release their own open-weight models, which anyone can download and modify, as a way to counterbalance China's growing presence. He's spoken with leaders at several AI labs in recent weeks.
Open-weight models from China are a separate worry here. Because they carry few or no safeguards, they can be downloaded by anyone and used to probe government systems for weaknesses. Our story on 20 July 2026, "China's Moonshot and Alibaba unveil giant AI models they say rival GPT-5 and Claude", reported that both labs planned to make their models free to download.
Should you worry?
The practical risk isn't that a foreign AI model will arrive in your inbox. It's that distillation lets a competitor close the capability gap quickly and cheaply, and that open-weight models with no guardrails make it easier for anyone, not just state actors, to probe critical systems. The administration's disagreement about the fix isn't reassuring: if Commerce thinks White House proposals are unworkable, whatever action comes next could be narrow, rushed or both.
The journalist who's followed this beat would tell you the real story isn't the headline split. It's that the White House announced Kimi K3 was built on a distilled Anthropic model on the same day it was still figuring out what to do about it.
