Anthropic has accused Alibaba of carrying out the largest attempt it has yet seen to extract the capabilities of its Claude AI models, in a letter to US senators that several outlets confirmed on June 24.
The letter, dated June 10 and addressed to Senator Tim Scott and Senator Elizabeth Warren of the Senate Banking Committee, says operators affiliated with Alibaba and its Qwen AI lab used roughly 25,000 fraudulent accounts to run 28.8 million exchanges with Claude between April 22 and June 5. Anthropic called it “the largest known distillation attack on Anthropic to date.” Alibaba’s shares fell about 3% on the news. Bloomberg first reported the letter.
Distillation is a training method in which a weaker model is built using the outputs of a stronger one, letting a developer approximate advanced capabilities without paying the full cost of training. Anthropic says the campaign targeted Claude’s most advanced abilities in coding, reasoning, autonomous agents and complex planning, and circumvented its rules barring access from China.
The company warned that such extraction could accelerate China’s frontier AI and that cloned models often strip out safety guardrails. It is worth noting the careful wording: Anthropic describes the accounts as affiliated with Alibaba and Qwen, which is not the same as proving Alibaba directed the effort, nor that its models successfully replicated Claude. Alibaba has not responded to the accusation.
The alleged campaign dwarfs earlier ones Anthropic has disclosed. In February it named three Chinese labs, saying DeepSeek ran more than 150,000 exchanges, Moonshot more than 3.4 million and MiniMax more than 13 million.
The Alibaba figure is more than double the largest of those. Anthropic used the letter to press lawmakers, asking them to let AI firms share threat intelligence on distillation without antitrust risk, to penalize unauthorized distillation, to tighten export controls and to strengthen intellectual-property protections for frontier systems. It argued that distillation converts hundreds of billions of dollars of US research into a subsidy for competitors.
The Policy Push
The accusation doubles as advocacy. Anthropic, valued at $965 billion and preparing for a public offering, has long lobbied for tighter controls on AI flowing to China, and the letter directly advances measures that would benefit US developers, including itself. That alignment of interest does not make the claims false, but it is part of how they should be read.
The letter also lands at a sensitive moment in Washington, sent ahead of a Senate hearing on AI and shared with White House officials, and it pointedly noted that Alibaba had been added to a Pentagon list of companies alleged to support China’s military. Alibaba is contesting that designation in court, calling it without “basis in fact or law.”
A Wider Standoff
The dispute sits inside a tense stretch for Anthropic itself. Earlier this month, the Trump administration ordered the company to suspend access to its latest models, Fable 5 and Mythos 5, for all foreign nationals, citing unspecified national security authorities, and Anthropic disabled them while senior staff traveled to Washington to negotiate.
The company says both sides are working to resolve it but has not said when the models will return. So Anthropic is simultaneously asking the government for stronger protection against foreign extraction and pushing back on government restrictions on its own technology. The episode underscores how AI has become a front in US-China competition, where model capabilities are treated as strategic assets and the lines between commercial rivalry, security policy and corporate lobbying increasingly blur.
