In a letter to two U.S. senators dated June 10, U.S. artificial intelligence company Anthropic accused Chinese company Alibaba of "illicitly" extracting the capabilities of its Claude AI model at the "largest" levels that the AI lab has measured, the Financial Times reported on June 24. Anthropic also argued for Congress to limit China's access to advanced U.S. computing power via export controls and clarify antitrust rules so U.S. AI labs can share information about detecting and preventing Chinese distillation attacks, by which labs issue high numbers of queries to AI models to analyze their responses and reasoning.

Anthropic said that between April 22 and June 5, Alibaba and Alibaba Qwen used nearly 25,000 fraudulent accounts to generate nearly 30 million exchanges, warning that Chinese distillation attacks can harvest U.S. intellectual property and accelerate China's timeline to develop capabilities on par with Anthropic's advanced Mythos model. However, Anthropic has not accused Z.ai — the developer of the GLM family of models, which are China's most robust in agentic AI and coding — suggesting that some of China's advancements in these areas are not solely due to distillation attacks.