AI company Anthropic accuses Alibaba of using thousands of fake accounts to copy Claude model

One of China’s leading artificial intelligence research companies, linked to Alibaba, is the target of a serious accusation. According to Anthropic, an American company, the Chinese institute would have carried out around 28.8 million “conversations” with its cutting-edge AI model to replicate it.
In a letter released on June 24, Anthropic told the United States Senate Banking Committee that an operator associated with Alibaba’s Qwen Research Institute opened approximately 25,000 fraudulent accounts. These accounts were allegedly used between April 22 and June 5 to interact millions of times with Anthropic’s Claude model, according to the complaint.
The fundamental purpose of this operation would be the “distillation” of the Claude model. This technique involves using a simpler AI model to interact with a more advanced model, collecting a vast volume of responses. Then the simple model is trained to emulate the results of the more sophisticated system. This allows you to create a functional replica of a cutting-edge model without the considerable development costs of the original.
It is crucial to emphasize that this is an accusation that has not yet been proven. Alibaba did not comment on specific points, and there was no verification by independent sources. However, if Anthropic’s claims are true, the method used to obtain a replica of a cutting-edge model on an industrial scale points to a strategy of rapid technological advancement.

Details of Anthropic’s communication to the US Senate
Anthropic forwarded its letter on June 10 to Senators Tim Scott and Elizabeth Warren, and the contents were made public about two weeks later. The document points out that this distillation operation against Claude would be the largest ever identified by the company. The vast amount of 25,000 accounts was designed to simulate normal traffic, avoiding speed limits and detection systems.
The fake accounts were not interested in Claude’s general conversational skills, but rather in specific, advanced skills. Among them, sophisticated software engineering and multi-stage agentive reasoning stand out, which enable models to plan and execute complex tasks autonomously. Anthropic framed the action as a national security risk, arguing that it is an attempt to convert billions of dollars in American investments in AI into subsidies for geopolitical competitors.






