The White House has accused three Chinese artificial intelligence firms of conducting large-scale theft of U.S. AI technology by illicitly extracting capabilities from Anthropic’s Claude model, according to a memo reviewed by BBC News.
The memo details industrial-scale distillation campaigns
The White House alleges that DeepSeek, Moonshot AI, and MiniMax created over 24,000 fraudulent accounts to interact with Claude approximately 16 million times, using a technique known as distillation to train their own competing models. Anthropic previously described similar activity earlier this year, identifying the same three Chinese labs as conducting coordinated efforts to copy its AI systems through unauthorized access to its chatbot outputs.
This follows a pattern of U.S. Accusations against Chinese AI firms
Last month, OpenAI made comparable claims to the U.S. House Select Committee on China, asserting that DeepSeek and other Chinese companies may have improperly distilled its ChatGPT models over the past year. The recurring allegations highlight ongoing tensions over intellectual property protection in the global AI race, particularly as U.S. Policymakers express concern about the effectiveness of export controls designed to limit China’s access to advanced semiconductor technology.
What is distillation in AI development?
Distillation is a common method in the AI industry where one model is trained on the outputs of another, often more capable system, to create smaller or more efficient versions. Whereas legitimate when done with proper authorization, Anthropic and OpenAI argue that using fraudulent accounts to access proprietary models violates terms of service and undermines U.S. Export controls aimed at constraining China’s access to cutting-edge AI technology.
How have the accused companies responded?
The source material does not include any public statements or responses from DeepSeek, Moonshot AI, or MiniMax regarding the White House memo or Anthropic’s earlier allegations of distillation campaigns.