US-China AI Policy Turf War: White House Debates Controls vs Open Models

The Trump administration is internally divided over how to handle China's rapidly advancing AI models, with the White House pushing for stricter controls on distillation and the Commerce Department favoring incentives for US open-weight models. For AI builders, this policy uncertainty could reshape how models are trained, published, and hosted.

What happened

Senior White House officials have discussed possible measures to stop Chinese AI labs from training models on US models through a practice called distillation, according to people familiar with the deliberations. The debate follows the release of Moonshot AI's Kimi K3 model, which rivals top US models from Anthropic and OpenAI. The White House announced that Moonshot developed Kimi K3 by distilling Anthropic's Claude Fable 5 model. Treasury Secretary Scott Bessent told Fox Business that covert distillation could lead to sanctions, describing it as "IP theft."

Anthropic last month alleged that Alibaba carried out the "largest known distillation attack" against it to date. The White House is expected to take some form of presidential action, though not necessarily an executive order. Meanwhile, the Commerce Department, which oversees export controls through the Bureau of Industry and Security (BIS), has viewed stricter restrictions as unworkable. Commerce Secretary Lutnick has contemplated creating incentives for top US labs to release their own open-weight models to counterbalance China.

Why AI builders should care

For developers and product teams building with or distributing AI models, the outcome of this policy debate could directly affect how open-weight models are regulated. If the White House pursues stricter controls, it may become harder to host or use Chinese open-weight models in US-based infrastructure. The Commerce Department last year considered adding multiple Chinese AI labs to its Entity List, which would cut off US access without a license. Such moves could fragment the model ecosystem and force builders to choose between compliance and access to cutting-edge models.

Open-weight models from China, which have few safeguards and can be downloaded by anyone, have escalated policy concerns about government system vulnerabilities. If the US imposes bans or liability requirements on hosting Chinese models, as Axios reported the White House considered, AI builders may need to implement stricter model provenance checks and compliance workflows.

Practical implications

Policy signals suggest two possible paths. One is stricter controls on distillation and training with US-origin data, which could limit how Chinese labs improve their models. The other is incentives for US labs to publish open-weight models, as Lutnick has discussed with AI lab leaders. For builders, the first path could reduce the availability of competitive open models from China, while the second could increase the supply of US open models with more safeguards.

If the US imposes sanctions for AI IP theft, as Bessent hinted, companies involved in distillation at scale could face legal risks. Builders should monitor whether their model training pipelines involve distillation from US models and assess compliance exposure. The Guardian reported that hawkish officials are rekindling efforts to block US access to Chinese AI models via de facto bans or executive orders.

Caveats

No formal White House action has been announced yet. All reported measures are still under internal discussion and could change as the administration debates the balance between security and innovation. The Commerce Department has not received a formal request for input from the White House. Builders should treat these signals as directional but not definitive, and avoid making irreversible infrastructure decisions based on unconfirmed policy shifts.