Anthropic Accuses Alibaba of Large-Scale Claude Model Distillation via Fake Accounts

Anthropic, the developer behind the Claude large language model, has brought to light what it describes as the largest known attack on its AI intellectual property, pointing fingers at Chinese e-commerce and technology behemoth Alibaba. According to a letter sent by Anthropic to the U.S. Senate Banking Committee, Alibaba's AI lab, Qwen, is alleged to have engaged in a systematic effort to "distill" Claude's advanced capabilities. This refers to a process where a less capable AI model is trained extensively on the responses and outputs generated by a more powerful model, effectively learning to mimic its performance. The alleged campaign reportedly took place between April 22 and June 5, 2026. During this period, Anthropic claims that operators affiliated with Alibaba established around 25,000 fraudulent accounts. These accounts were then used to conduct an staggering 28.8 million exchanges with the Claude AI model. The sheer scale of this operation suggests a deliberate and industrial-level attempt to extract valuable insights and functionalities from Claude. Anthropic's letter, dated June 10, was made public approximately two weeks later and was addressed to Senators Tim Scott and Elizabeth Warren, who serve as the chair and ranking member of the Senate Banking Committee, respectively. The company emphasized that this distillation effort was not merely about casual interaction but specifically targeted Claude's cutting-edge features, such as advanced software engineering capabilities and its multi-step agentic reasoning, which allows the model to plan and execute complex tasks autonomously. The incident underscores the growing concerns surrounding intellectual property theft and competitive espionage within the rapidly evolving artificial intelligence sector. Anthropic views this as a clear attempt to accelerate China's access to advanced AI capabilities, specifically referencing Anthropic's 'Mythos Preview' technology. The company has expressed its support for U.S. government initiatives aimed at combating such attacks, including collaborative efforts with private sector AI firms through threat intelligence sharing. This accusation follows previous reports from Anthropic in February regarding similar, albeit smaller, distillation campaigns by other Chinese AI labs, indicating a pattern of escalating challenges in protecting proprietary AI models.