White House draws new AI line on China

The Trump administration this week drew a sharp new line in its approach to artificial intelligence, accusing a Chinese company of stealing U.S. technology through a common AI practice called distillation, even as it reaffirmed support for the open development of AI models. The distinction could become the foundation for U.S. action against Chinese AI companies without broadly restricting the open-weight models that businesses, researchers and startups increasingly rely on.

What Is Model Distillation?

Distillation is a technique where a smaller "student" model learns from a larger "teacher" model, replicating its behavior without undergoing the same extensive training. This process produces compact models that run more efficiently and cheaply while maintaining similar performance. It is widely used in the AI industry to create specialized or cost-effective versions of powerful models. In the administration's view, legitimate small-scale or authorized distillation is a vital part of the innovation ecosystem. However, the White House is now drawing a clear line against what it calls "large-scale, covert industrial distillation" aimed at stealing proprietary U.S. technology.

Accusations Against Chinese Firm Moonshot

The administration specifically accused Beijing-based Moonshot AI of using unauthorized distillation to copy Anthropic's Fable model. Michael Kratsios, President Trump's chief technology adviser, said on X that while legitimate distillation plays a vital role, industrial-scale theft conducted through fraudulent accounts or evasive access techniques is unacceptable. Other top officials, including Treasury Secretary Scott Bessent and U.S. Trade Representative Jamieson Greer, echoed similar claims of intellectual property theft. The accusations come amid growing concerns over Moonshot's Kimi K3 model, which offers competitive performance at a fraction of the cost of leading U.S. counterparts. Some U.S. companies have begun opting for these cheaper alternatives, rattling policymakers and industry players alike.

Industry Reactions and a Joint Evaluation

OpenAI co-founder Greg Brockman told Bloomberg it was too early to determine if recent Chinese models had been distilled from GPT systems, and he said he had not discussed banning Chinese open models with the administration. He described distillation as "definitely a technical issue," suggesting that the problem may be more amenable to technical solutions than policy action. In a joint evaluation released Thursday, the U.K. Artificial Intelligence Security Institute and the U.S. Center for AI Standards and Innovation assessed Kimi K3's cyber capabilities and found it performs significantly below other frontier models. That finding could influence how seriously the threat is taken.

Looking Ahead: A New Framework for AI Competition

The administration's comments suggest it is laying the groundwork for targeted action against potential AI IP theft via distillation, while preserving the open ecosystem that has fueled U.S. leadership in AI. By defining two categories of distillation—one legitimate and one theft—the White House hopes to prevent Chinese companies from free-riding on American innovation without stifling the open-weight models that benefit the global AI community. As the administration weighs next steps, all eyes will be on whether it imposes new restrictions on Chinese AI firms or works with industry to develop technical safeguards against unauthorized distillation.