White House Weighs New Response as China’s AI Models Near US Front-Runners

The Trump administration is split over how to respond to China’s accelerating AI progress, with the White House pushing for tougher controls and the Commerce Department arguing that some of the proposed restrictions may be impractical. The dispute matters because Chinese labs are now producing models that can compete with leading US systems, raising fresh fears about intellectual property theft, cyber risk, and the limits of export controls.

Over the past week, that tension has sharpened around Moonshot AI’s Kimi K3 model and allegations that Chinese companies are using “distillation” techniques to copy the capabilities of frontier American systems. The White House is now weighing a possible presidential action, while officials inside Commerce are said to be looking for more workable ways to respond.

What is driving the White House debate?

The core issue is whether the United States can stop Chinese AI developers from learning from powerful American models without imposing rules that are too broad to enforce. Senior officials have been discussing how to limit model distillation, a process in which one AI system is trained on the outputs of another to reproduce similar performance at lower cost.

That concern has become more urgent after a series of revelations and announcements tied to Chinese AI labs. Anthropic recently alleged that Alibaba had carried out what it described as the largest known distillation attack against its own systems. Then, the administration said Moonshot AI had used Anthropic’s Claude Fable 5 model as part of the development process behind Kimi K3.

Those developments have turned a technical debate into a national-security and trade-policy fight. In Washington, the question is no longer simply whether Chinese models are catching up. It is whether the United States can prevent them from doing so by borrowing from US breakthroughs.

Why distillation has become the focal point

Distillation is attractive to model builders because it can speed development, reduce costs, and help smaller teams imitate the strengths of bigger systems. But from the perspective of US officials and AI companies, it can also look like a shortcut around expensive research, one that may let foreign competitors benefit from American investment without permission.

The administration is treating the issue especially seriously because the technique may be difficult to police at scale. Unlike a conventional hack, distillation can be hidden in ordinary model training workflows, making it harder to prove and easier to deny.

“IP theft” is how Treasury Secretary Scott Bessent characterized covert distillation in remarks to Fox Business, suggesting the government could eventually consider sanctions in response.

That language underscores how quickly the issue has moved from industry concern to a potential sanctions and export-control matter. The White House is not merely looking at technology competition; it is considering whether American AI models are now a target of systematic appropriation.

How does distillation differ from a traditional cyberattack?

It does not necessarily involve breaking into a system, which is why it can sit in a legal and policy gray zone. Instead, it relies on repeated queries, observation of outputs, and training methods that let a new model learn from a better one’s behavior. That makes it harder to detect and even harder to regulate cleanly.

What happened with Moonshot AI and Kimi K3?

Moonshot AI’s release of Kimi K3 last week intensified the policy fight because the model reportedly rivals some of the most capable systems from Anthropic and OpenAI. The administration subsequently said Moonshot had built Kimi K3 by distilling Anthropic’s Claude Fable 5 model.

That announcement gave new urgency to an argument already underway in Washington: if Chinese labs can replicate American AI performance by studying outputs from leading US models, then current safeguards may not be enough. The Kimi K3 episode also suggested that Chinese developers are not just keeping pace, but finding ways to exploit the very systems meant to keep the United States ahead.

The story is especially sensitive because frontier AI is now central to both commercial competition and government security planning. As models become more capable, the stakes increase for law enforcement, defense, critical infrastructure, and intelligence use cases.

How the administration is split

The disagreement inside the administration is less about whether China’s AI rise matters than about how far Washington should go to stop it. On one side, the White House has backed stricter controls aimed at slowing Chinese labs that may soon match the best American models. On the other, Commerce officials have warned that some of those ideas may be difficult to implement or too blunt to achieve their goals.

Commerce’s role is especially important because it oversees export controls through the Bureau of Industry and Security. That gives the department practical leverage over policy, even if the White House sets the political direction.

According to people familiar with the discussions, no formal request for interagency input has yet been sent to Commerce, even though proposals are already circulating. That suggests the administration is still trying to settle on a policy frame before moving into the more bureaucratic process of drafting rules or orders.

What is Commerce proposing instead?

Commerce has not publicly detailed its preferred answer, but officials there are said to believe that some of the White House’s ideas are not realistic. The department appears to favor solutions that can be enforced more effectively and that do not create a patchwork of rules impossible to monitor across the global AI ecosystem.

One concept reportedly under discussion is creating incentives for major US labs to develop open-weight models of their own. The idea would be to counter China’s open-weight releases by giving American companies a way to compete in an ecosystem where model weights are downloadable and usable by anyone.

That approach reflects a more mixed strategy: not only restricting rivals, but also encouraging domestic innovation in formats that can compete directly with Chinese offerings.

Why open-weight models are changing the policy fight

Open-weight models have become a major policy headache because they lower the barrier to deployment and use. Once weights are available, developers can fine-tune, modify, and run the model without depending on the original company’s hosted service.

For policymakers, the downside is obvious. Models with little or no built-in safeguard can be repurposed for harmful uses, including attempts to probe or attack government systems. The attraction for researchers and developers is also obvious: open models spread faster, invite experimentation, and can accelerate innovation.

China’s growing use of open-weight releases has forced Washington to confront a new dilemma. Tight rules may slow adversaries, but they may also leave US companies and researchers at a disadvantage if they cannot freely share or adapt their own systems.

How are open-weight models different from closed models?

Open-weight models expose the trained parameters of the system, allowing others to download and reuse them. Closed models, by contrast, are usually accessible only through an API or a hosted product, which gives the company more control over how they are used and monitored.

• Open-weight models: Easier to customize, more difficult to police.
• Closed models: More controlled, but less flexible for users.
• Policy challenge: Open access can boost innovation while increasing abuse risk.

What role is Anthropic playing in the dispute?

Anthropic has become one of the most influential voices in the debate because it is both a competitor in frontier AI and a complainant about alleged misuse of its technology. Its public allegations about Alibaba’s use of distillation have helped focus White House attention on the broader risk of model copying.

The company is also central to the government’s latest concern about Moonshot AI. By linking Kimi K3 to Claude Fable 5, the administration effectively turned a product release into a policy test case.

Earlier in the month, Anthropic’s Claude Mythos and Fable 5 systems were already a source of administration anxiety for a different reason: their ability to discover vulnerabilities in government systems and critical infrastructure. That meant the same models were seen both as strategic assets and as possible tools for identifying security weaknesses.

Anthropic’s allegations and the administration’s follow-up show how frontier model releases can quickly become a matter of national security, not just competition among tech companies.

Why the Commerce Department matters more than ever

Commerce has emerged as a crucial player because it administers export controls, the government’s main tool for limiting access to sensitive technologies. In the AI era, those controls are no longer about chips alone; they also shape who can build, train, and deploy advanced models.

That gives the department an unusual position inside the administration’s internal debate. The White House may want stronger action, but Commerce must decide what can realistically be enforced without creating loopholes or collateral damage to US firms.

People familiar with the discussions say Commerce Secretary Howard Lutnick has spoken with leaders at several AI labs in recent weeks. That suggests the department is trying to gather industry feedback before settling on a policy direction.

Lutnick also appears to be navigating a middle course. According to the source material, he has imposed export controls on Anthropic to rein in the company when needed, but he has also been more flexible than some other officials. That balancing act reflects the broader difficulty of writing AI policy in real time.

How did the issue evolve so quickly?

The debate escalated because the technical and geopolitical landscape changed faster than US policy could adapt. Just weeks ago, the dominant concern inside the administration was whether Anthropic’s models were becoming capable enough to uncover vulnerabilities in sensitive systems. Now, the spotlight has shifted to China’s ability to release near-frontier models that may have been built partly through distillation.

This is a classic policy whiplash problem. By the time Washington identifies one risk, the market and the research community may have already moved on to another.

China’s speed is what makes the current moment particularly volatile. If open-weight models can be downloaded and reworked immediately, then policy responses built around older assumptions about closed labs and slow diffusion may not hold up for long.

Timeline: how the AI policy fight escalated

What options does the White House appear to be considering?

The White House is expected to take some form of presidential action, though a formal executive order is not currently anticipated. That leaves a range of possibilities, including sanctions, guidance to agencies, new enforcement priorities, or other direct steps that do not require the same procedural weight as an order.

Officials are clearly trying to balance speed against durability. A quick response may signal resolve, but a poorly designed one could be easy to evade or could hurt American firms more than Chinese competitors.

Because the details remain fluid, the ultimate policy response may be narrower than some hawks want but more forceful than Commerce would prefer. For now, the administration appears to be testing where that line should sit.

Why this matters beyond Washington

The dispute is about more than one model or one company. It speaks to the future of global AI competition, the fragility of intellectual property in the model era, and the limits of government control over a technology that can be copied, adapted, and distributed at high speed.

If the US moves too aggressively, it risks discouraging collaboration and innovation among domestic firms. If it moves too cautiously, it could allow Chinese labs to close the capability gap using American research as a training wheel.

The policy challenge is especially acute because AI is no longer just a consumer product or research frontier. It is increasingly part of cybersecurity, infrastructure protection, intelligence analysis, software development, and military planning. That makes every model release a potential strategic event.

For now, the administration’s internal split reflects that reality. The White House wants to slow China down. Commerce wants rules that actually work. And the speed of AI development is leaving both sides with little time to decide.

What comes next?

The next step is likely a White House decision on whether to announce new measures, direct agencies to study the issue, or signal a tougher enforcement posture. Any move will probably be watched closely by AI companies in the United States and abroad, especially those building models that could be vulnerable to imitation.

Industry executives will also be looking for clues about whether Washington intends to encourage more open-weight releases from US labs. That would be a notable shift, because it would mean the government is not only trying to block Chinese progress, but also trying to compete with it on the same distribution model.

For now, the most important fact is that the administration sees the problem as urgent enough to merit presidential attention. With China’s leading models improving rapidly and accusations of distillation multiplying, the debate in Washington is moving from theory to policy.

Key points at a glance

• The Trump administration is divided over how to respond to China’s rapid AI progress.
• The White House wants tougher restrictions on model distillation and related practices.
• The Commerce Department believes some of those ideas may be hard to enforce.
• Moonshot AI’s Kimi K3 and Anthropic’s allegations against Alibaba have intensified the debate.
• A presidential response is expected, though not necessarily an executive order.

As the policy fight continues, one question now dominates: can the United States slow China’s AI rise without undermining its own innovation engine? The answer may determine not just the next round of export controls, but the broader balance of power in artificial intelligence.