Anthropic has publicly disclosed that three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — conducted industrial-scale model distillation against Claude using more than 24,000 fake accounts and over 16 million interactions. The operation was designed to extract Claude's capabilities and transfer them into the labs' own models.
The distillation campaigns were targeted. DeepSeek focused on reasoning capabilities, running millions of complex reasoning queries designed to capture Claude's chain-of-thought patterns and logical inference abilities. MiniMax targeted agentic coding specifically, with approximately 13 million exchanges aimed at replicating Claude's ability to plan, execute, and debug multi-step coding tasks autonomously.
The disclosure comes at a politically sensitive moment, with US policymakers actively debating AI chip export controls to China. Anthropic's decision to go public appears calculated to influence that debate by demonstrating that Chinese labs are not just building their own models — they are systematically extracting capabilities from American AI systems. The scale of the operation, spanning tens of thousands of accounts and millions of interactions, suggests organised institutional effort rather than individual researchers.
