Anthropic disclosed that Alibaba used 25,000 fake accounts and 28.8 million API exchanges during April-June 2026 to illicitly distill Claude model for training Alibaba's Qwen AI.

Anthropic, the American AI lab behind one of the most advanced large language models available today, has claimed in a letter to the U.S. Senate that Chinese tech giant Alibaba has illicitly used Claude to train its own models. According to Reuters, the company sent the letter to Sen. Tim Scott (R-S.C.), chair of the U.S. Senate Banking Committee, and Sen. Elizabeth Warren (D-Mass.), the ranking member, ahead of a scheduled hearing on AI issues.

This is not the first time Anthropic has accused Chinese AI labs of "stealing" Claude's capabilities to train competing models. Earlier this year, the company claimed that DeepSeek, Moonshot, and MiniMax collectively used 24,000 fraudulent accounts and made 16 million exchanges to train their own large language models on Claude's output. The technique in question, called distillation, involves training an AI system using the output of a more advanced model. While legitimate applications exist—such as distilling a frontier AI model to create a lighter, cheaper version of itself—Anthropic argues that competing labs exploit the same technique to build their own models "at a fraction of the time, and at a fraction of the cost, that it would take to develop them independently."

Anthropic says it traced the distillation campaign back to operators with connections to Alibaba, one of China's largest tech companies often compared to Amazon, and Alibaba Qwen, its AI research division. The lab warns that distillation could enable China to develop a frontier AI model capable of matching the capabilities of advanced American systems—a prospect that concerns many U.S. lawmakers.

While American tech companies maintain an edge in cutting-edge AI models, Chinese competitors are advancing rapidly. Elon Musk has estimated that a Chinese AI lab could develop a Fable 5-class model by the first quarter of next year, though the CEO of Chinese AI startup Z.ai confidently responded that "won't take that long." Many enterprise users are increasingly adopting more affordable open-source Chinese models for their agents as token costs surge, reserving the most powerful—and expensive—American models for complex tasks only.

Both nations are intensifying efforts to achieve AI supremacy, with the U.S. imposing export controls to limit Chinese access to advanced hardware for chipmaking and AI training, while China has countered with its own restrictions on rare earth materials essential to chip manufacturing.