U.S. Targets Chinese AI 'Distillation' Campaigns

The Trump administration has announced a crackdown on foreign entities "principally based in China" accused of illegally extracting capabilities from leading American AI systems.

The initiative, outlined in a memo from chief science and technology adviser Michael Kratsios, directs U.S. AI companies to help identify these activities, build defenses, and develop methods to punish offenders.

A Narrowing Gap

The memo cites a Stanford University report showing the U.S.-China gap in AI model performance has "effectively closed." This narrowing competitive landscape forms the backdrop for the administration's heightened concern over intellectual property theft.

Accusations and Denials

Michael Kratsios accused foreign entities of "exploiting American expertise and innovation."

"Model extraction attacks are the latest frontier of Chinese economic coercion and theft of U.S. intellectual property." — Rep. Bill Huizenga (R-MI), sponsor of related House legislation

David Sacks, former AI and crypto adviser to President Trump, alleged that Chinese startup DeepSeek "distilled the knowledge out of OpenAI's models." OpenAI made similar claims in a February letter to lawmakers, and Anthropic accused DeepSeek and two other China-based labs of campaigns to "illicitly extract Claude's capabilities."

China's embassy in Washington pushed back. Spokesperson Liu Pengyu opposed "the unjustified suppression of Chinese companies by the U.S." and affirmed that China remains committed to intellectual property protection.

The DeepSeek Factor

The controversy gained momentum after Chinese startup DeepSeek released a low-cost large language model in 2024 that proved competitive with U.S. counterparts. This breakthrough drew scrutiny to the practice of model distillation—a legitimate AI training technique that becomes problematic when competitors use it to rapidly acquire capabilities without independent development.

Legislative Action

The House Foreign Affairs Committee has advanced a bill targeting foreign extraction of "key technical features" of U.S.-owned AI models, with sanctions as a potential punishment.

Acknowledged Use of Chinese Models

In a notable twist, San Francisco-based Anysphere acknowledged that its latest product was based on an open-source model by Chinese company Moonshot AI, highlighting the cross-border nature of AI development.

Expert Perspective

Kyle Chan of the Brookings Institution noted the practical challenges:

Distinguishing unauthorized distillation from legitimate data requests is difficult, but information sharing and government facilitation could help.

The administration's strategy will rely on collaboration with American AI companies to identify violations and build defensive measures against future extraction attempts.