White House Accuses China of Industrial-Scale AI Model Distillation
The US has accused China of "industrial-scale" copying of its artificial intelligence models, a practice known as distillation. According to a policy memo released by the White House Office of Science and Technology Policy (OSTP), the US government plans to share intelligence with American AI companies about these foreign distillation campaigns and explore accountability measures.
Michael Kratsios, director of OSTP, stated: "We have evidence that foreign entities, primarily in China, are running industrial-scale distillation campaigns to steal American AI. We will be taking action to protect American innovation." This announcement comes just three weeks before a highly anticipated Trump-Xi summit on May 14th.
Distillation Explained
Distillation is a technique where a cheaper model approximates the capabilities of a more advanced, "frontier" AI model by learning from its responses to carefully constructed queries. While legal questions surround this practice, its strategic implications are clear.
Evidence and Allegations
The OSTP memo supports allegations made by US AI companies since February. OpenAI accused DeepSeek of distilling its models, providing evidence of accounts associated with DeepSeek employees bypassing access restrictions to extract model outputs at scale.
Anthropic, another AI company, published detailed evidence on February 23rd, naming three Chinese labs—DeepSeek, MiniMax, and Moonshot AI. They claimed that DeepSeek conducted over 150,000 exchanges with Claude, a large language model developed by Anthropic, focusing on foundational logic and alignment techniques.
The Deterring American AI Model Theft Act
The memo's release coincides with the introduction of the Deterring American AI Model Theft Act (H.R. 8283) on April 15th, highlighting the urgency of the issue.