Anthropic, the U.S. company behind the Claude AI assistant, has accused Chinese tech giant Alibaba of running a large-scale, illicit campaign to extract capabilities from its models. According to a letter reviewed by Reuters, Anthropic sent its allegations to U.S. senators Elizabeth Warren and Tim Scott.
The numbers are striking. Anthropic says the effort spanned roughly 25,000 accounts and 28.8 million exchanges, or conversations, on Claude. Tom's Hardware reports the activity ran from April to June 2026, and that Anthropic traced the accounts to operators affiliated with Alibaba and its Qwen AI division. Several outlets describe the alleged use of fake accounts to avoid detection.
At the heart of the accusation is a technique known as "distillation." In plain terms, that means using one AI model's outputs to train a rival model, in effect copying its hard-won abilities without building them from scratch. Anthropic frames this as systematic data theft aimed at training a competitor; marketscreener.com and others quote it characterizing the campaign as "brazen."
The fallout is already showing up in markets. Benzinga and MSN report that Alibaba stock (ticker BABA) was in the spotlight as the news spread. There is also an awkward backdrop: Music Business Worldwide notes that Anthropic is itself fighting lawsuits over the alleged copying of song lyrics to train Claude, even as it now accuses Alibaba of copying its work.
The sources here are largely news aggregations and reports of Anthropic's letter; Alibaba's response, if any, is not detailed in them.
Why it matters: this is a flashpoint in the U.S.-China AI race, where the question of who can legitimately learn from whom is increasingly fought over in courtrooms and lobbying letters rather than labs.