Anthropic has accused operators affiliated with Alibaba’s Qwen AI lab of conducting the largest known model distillation campaign ever launched against the company, according to a letter the US AI safety firm sent to the Senate Banking Committee on June 10.
What happened
Between April 22 and June 5, 2026, nearly 25,000 fraudulent accounts generated 28.8 million exchanges with Claude, according to Anthropic’s account. The campaign targeted Claude’s software-engineering and agentic-reasoning capabilities — specifically the Mythos Preview model, Anthropic’s most advanced frontier system.
The technique is called knowledge distillation: a less capable model is trained on the outputs of a more powerful one, letting a rival replicate some of its capabilities without building them independently. Anthropic says the operation was designed to close the gap between Alibaba’s Qwen model and Claude Mythos Preview.
Congressional response
Anthropoc addressed the findings to Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren, calling for stronger legislative tools to counter such campaigns. The disclosure has already prompted action: Senators Bill Hagerty (R-TN) and Andy Kim (D-NJ) are working to add an amendment to defense legislation that would blacklist or sanction entities found conducting AI distillation attacks against US companies.
Not the first allegation of its kind
This is not the first time Anthropic has raised alarms about Chinese AI companies extracting capabilities from Claude. In February 2026, the company reported similar campaigns by DeepSeek, Moonshot AI, and MiniMax, ranging from roughly 150,000 exchanges attributed to DeepSeek to more than 13 million by MiniMax. Anthropic said at the time that such campaigns were growing in scale and sophistication.
Alibaba declined to comment when contacted by reporters.