According to a copy of a letter obtained by Bloomberg, AI company Anthropic has contacted multiple U.S. senators and White House officials, accusing entities linked to Alibaba's Qwen lab of launching 28.8 million conversations against its Claude model between April and June this year using nearly 25,000 fake accounts. The attacks were not random but precisely targeted Claude's two most competitive capabilities: software engineering and agentic reasoning.
Adversarial Distillation: Industrial-Scale Capability Theft
Anthropic characterized the operation as an "adversarial distillation attack." The strategy involved using bulk fake accounts to query Claude, collect responses, and then feed that data into training Alibaba's own Qwen models. This allowed the latter to replicate the reasoning abilities of American frontier models without bearing the multi-billion-dollar R&D costs. Anthropic noted that the method is identical to what it previously identified in DeepSeek and MiniMax — systematic, large-scale, and industrial.
Industry Defense and Legal Gray Areas
Anthropic, OpenAI, and Google have now formed a three-way alliance to share information on detected distillation violations, turning individual concerns into collective defense. However, model distillation itself is not illegal — it is a common technique in AI research where small models learn from larger ones. The controversy lies in the scale and intent. When tens of thousands of fake accounts are used to industrial-scale extract frontier model capabilities, it violates the terms of service of major AI labs. Interestingly, Anthropic acknowledged in its letter that it also used distillation of its own earlier models when training Claude. The legal line between "distilling oneself" and "distilling others" remains unclear, and Anthropic has asked the U.S. government to clarify antitrust guidelines to allow freer sharing of distillation threat data among American companies.
U.S. Congress Legislates
Lawmakers in both chambers of the U.S. Congress are now moving. In the Senate, Republican Bill Hagerty and Democrat Andy Kim plan to introduce an amendment to blacklist or sanction Chinese companies that illegally distill U.S. AI outputs. In the House, Bill Huizenga and Sydney Kamlager-Dove are advancing a similar bipartisan bill, with both measures expected to be folded into the annual National Defense Authorization Act. This incident marks a shift in U.S.-China AI competition from model capability showcases to legal and regulatory confrontations.
(Based on Bloomberg reporting and Anthropic's official letter; some background from industry analysis.)