Anthropic has accused Chinese technology company Alibaba of attempting to copy capabilities from its artificial intelligence systems through what the U.S. AI firm described as a large-scale “distillation” effort, according to a letter sent to the Senate Committee on Banking, Housing, and Urban Affairs and confirmed by CNBC.
The June 10 letter, addressed to Republican Sen. Tim Scott and Democratic Sen. Elizabeth Warren, said operators linked to Alibaba and its AI lab conducted 28.8 million interactions with Anthropic models using about 25,000 allegedly fraudulent accounts between April 22 and June 5. Anthropic called the activity the largest known distillation attack against the company.
Distillation is a technique in which outputs from a more capable AI model are used to train a smaller or less advanced model. AI developers have raised concerns that the method can be used to replicate proprietary systems without authorization.
Anthropic said it is seeking closer coordination between industry and government to address illicit distillation and protect U.S. leadership in AI. Alibaba did not immediately respond to CNBC’s request for comment. Bloomberg first reported the letter.
The allegations come after an April White House Office of Science and Technology Policy memorandum pledged support for AI companies working to detect and respond to industrial-scale distillation. Anthropic said Alibaba proceeded despite administration warnings.
The company has previously reported similar campaigns involving DeepSeek, Moonshot and MiniMax, saying in February that such efforts were becoming more sophisticated.
At the same time, Anthropic is dealing with a separate dispute with the Trump administration over export controls. Earlier this month, the company said it was ordered to suspend access to its newest Claude models, Fable 5 and Mythos 5, for foreign nationals, including some employees. Anthropic said senior staff have been meeting with officials to resolve the matter. No timeline has been announced publicly.