Anthropic says Alibaba ran a large-scale effort to extract advanced capabilities from its Claude AI model, using tens of thousands of fake accounts and nearly 29 million prompts. The claims come from an internal letter obtained by Reuters, which Anthropic says describes the biggest known distillation attack against its systems so far.
According to the company, between April 22 and June 5, 2026, more than 28.8 million interactions were carried out with Claude through nearly 25,000 fraudulently created accounts. Anthropic believes the goal was to harvest large amounts of answers and data from the model and use them to train competing systems. The company said the operation was aimed at helping Chinese AI development get closer to the capabilities of Mythos Preview, one of its most advanced models.
Anthropic said the campaign was linked to operators associated with Alibaba and its AI lab, Qwen. If the allegations are confirmed, they would amount to one of the most direct and severe accusations yet made against a Chinese technology company in artificial intelligence. At the time Reuters reported the story, Alibaba had not issued a formal response.
The case follows similar complaints Anthropic made in February, when it said it had detected attempts by several Chinese AI labs to siphon capabilities from Claude. It then cited more than 150,000 suspicious interactions by DeepSeek, more than 3.4 million by Moonshot AI, and over 13 million by MiniMax. Anthropic warned at the time that the problem was becoming more sophisticated and aggressive, and required cooperation between tech companies, governments, and security agencies. The dispute comes amid intensifying U.S.-China AI rivalry. In April, the White House accused China of systematic theft of intellectual property from American AI labs, while the Pentagon recently added Alibaba to a list of Chinese firms with possible ties to the Chinese military. The U.S. Commerce Department has so far avoided trade restrictions on DeepSeek, despite security concerns, to avoid escalating tensions with Beijing. Two days after Anthropic sent its Senate letter, the U.S. Commerce Department imposed restrictions on Anthropic’s new Mythos and Fable models, citing fears that hostile intelligence and military actors could exploit them, forcing Anthropic to disable worldwide access.
