Anthropic has accused Alibaba of running a massive, coordinated operation to extract the capabilities of its Claude AI model without permission. The U.S. AI company calls it the largest known attack of this kind it has ever faced, and the accusation lands at a moment when tensions between Washington and Beijing over AI technology are already running high.
According to Reuters, Anthropic laid out the allegations in a letter dated June 10, sent to Senators Tim Scott and Elizabeth Warren, the chair and ranking member of the U.S. Senate Banking Committee, ahead of a scheduled hearing on AI. Alibaba has not responded to requests for comment.
The letter describes what Anthropic calls a “distillation” attack, a technique where a weaker AI model is trained on the outputs of a stronger one. In plain terms, it is a way to copy the intelligence of a more advanced system without building it from scratch.
The scale of the alleged operation is striking. Anthropic says it ran between April 22 and June 5, 2026, and involved:
- More than 28.8 million exchanges with Claude
- Almost 25,000 fraudulent accounts
- Operators affiliated with Alibaba and its AI lab, Alibaba Qwen
Anthropic says the goal was to help China replicate the capabilities of its most advanced model, Mythos Preview, at a faster pace than would otherwise be possible.
This is not the first time Anthropic has raised the alarm over Chinese AI companies allegedly stealing its technology. In February, the company disclosed that three Chinese AI labs had run similar operations against Claude. DeepSeek, the startup whose cheap AI model rattled Wall Street in January 2025, was involved in an operation that generated over 150,000 exchanges. Moonshot AI ran one at a scale of over 3.4 million, and MiniMax exceeded 13 million. At the time, Anthropic warned that these campaigns were growing in “intensity and sophistication.”
The Alibaba accusation dwarfs all of those combined, which puts the latest allegation in a different category entirely. It also comes as Alibaba faces separate pressure in Washington. The company was added to the Pentagon’s list of Chinese military companies earlier this month, a designation it is contesting.
The broader context matters here. In April, the White House accused China of stealing U.S. AI intellectual property on an industrial scale. Anthropic says in the letter that it supports U.S. government efforts to fight these attacks, including through threat-intelligence sharing between government and private AI companies.
The timing of events around the letter is also worth noting. On June 12, just two days after Anthropic sent it to Congress, the U.S. Commerce Department imposed restrictions on Anthropic’s Mythos and Fable AI models. Officials said they feared the models could be used by military intelligence in China and other countries of concern. The restrictions were sweeping enough that Anthropic had to disable access to the models globally.
Meanwhile, the Commerce Department has held off placing DeepSeek on a trade blacklist, despite an interagency government committee flagging it as a national security risk. Reuters reported this month that the department is trying to avoid further escalating tensions with Beijing, which shows just how complicated the policy response to AI theft has become. Naming the problem is one thing. Deciding what to do about it is another.
