Artificial intelligence developer Anthropic has accused operators associated with Alibaba and its Qwen laboratory of conducting the largest distillation campaign in history for the Claude model. In a letter to US Congressional Senators, the company reported that from late April to early June, more than 28.8 million interactions with the network were made through nearly 25,000 fake accounts. The main targets of the attack were Claude’s ability to write code, perform agent tasks and make long-term plans.

Anthropic emphasized that such actions allow Chinese laboratories to benefit from billions of US investments in AI without incurring the costs of training their own models. To protect its developments, the company called on legislators to expand intelligence sharing between AI developers and the government, clarify antitrust rules, and tighten export controls on advanced AI chips, closing loopholes for access to foreign data centers.

This is not the first incident of its kind: Anthropic had previously blamed DeepSeek, Moonshot AI and MiniMax for large-scale distillation. The situation is exacerbating geopolitical tensions in technology. Amid these developments, US Congressmen have already begun to promote legislation that would sanction foreign entities illegally extracting characteristics from closed US AI models, and Nvidia is developing technologies to track the smuggling of its accelerators into China.