White House Divided Over How to Counter the Rise of Chinese AI

The Trump administration is internally split on how to respond to China's rapidly advancing AI models, with the White House pushing for tighter controls and the Commerce Department skeptical of their workability. Distillation attacks — where Chinese labs train on US models — have become the central flashpoint.

A quiet but consequential policy battle is unfolding inside the Trump administration over how — or whether — to restrain China's surging AI capabilities, according to multiple people familiar with the deliberations reported by Wired.

The Core Divide

The split runs roughly between the White House, which is pushing for stricter measures against Chinese AI labs, and the Commerce Department, which views many of those proposals as unworkable in practice. The Commerce Department's influence stems largely from its oversight of export controls through the Bureau of Industry and Security (BIS).

The debate has sharpened in recent weeks following two significant developments:

Moonshot AI, a Chinese lab, released its Kimi K3 model, which the White House claims was built by distilling Anthropic's Claude Fable 5 — and which reportedly rivals the top US frontier models from Anthropic and OpenAI.
Anthropic separately alleged that Chinese tech giant Alibaba had carried out what it called the "largest known distillation attack" against it to date.

What Is Distillation — and Why Does It Matter?

Model distillation is the practice of training a smaller or newer AI model on the outputs of a more powerful one. It's a legitimate technique in AI research, but it becomes contentious — and potentially illegal — when done covertly against a competitor's proprietary model without authorization.

The White House is treating covert distillation with particular urgency. Treasury Secretary Scott Bessent told Fox Business that the practice amounted to "IP theft" and suggested it could lead to sanctions. Senior White House officials have been actively discussing measures to prevent Chinese labs from exploiting US models this way.

Presidential action of some kind is expected in response — though one source indicated it is not expected to take the form of an executive order.

Commerce Takes a Different View

Commerce Secretary Howard Lutnick appears to be navigating a middle path. Rather than blanket restrictions, he has been exploring incentives for top US labs to release their own open-weight models as a strategic counterbalance to China's growing open model ecosystem. Lutnick has reportedly spoken with leaders across several AI labs in recent weeks.

Notably, no formal request for input has yet been sent from the White House to the Commerce Department — suggesting the administration is still in the early stages of forming a coherent policy response.

The Open-Weight Problem

Underlying all of this is a deeper concern about open-weight Chinese AI models — models with few or no safety guardrails that can be freely downloaded and deployed by anyone. Unlike API-gated commercial models, open-weight releases cannot be revoked or restricted once they're out.

For US policymakers, these models present a dual threat: they can be used to probe vulnerabilities in government systems and critical infrastructure, and they erode the leverage that export controls on chips and model access were designed to create.

Earlier this month, the administration's primary AI concern was Anthropic's own frontier models — specifically Claude Mythos and Fable 5 — because of their ability to identify security vulnerabilities. That Chinese labs may now be building on top of those same capabilities, covertly and at scale, has significantly raised the stakes.

What This Means for AI Startups and Founders

For founders building on top of frontier models — particularly those working with API access from Anthropic or OpenAI — this policy moment is worth watching closely:

Terms of service enforcement around distillation is likely to tighten, with potential legal and regulatory teeth attached.
Export controls could expand to cover model weights or fine-tuning access, affecting how international teams build and deploy AI products.
The push to release US-origin open-weight models as a geopolitical counterweight could reshape the open-source AI landscape — creating new base models for startups to build on.

The administration hasn't landed on a policy yet, but the direction of travel is clear: AI model access is becoming a national security issue, and the rules of the road are about to get more complicated.