Anthropic accuses Alibaba of a large-scale effort to extract its Claude models
Anthropic has told the White House that Alibaba and its AI lab “brazenly” and “illicitly” tried to extract the capabilities of its Claude models, describing the largest distillation campaign it has detected to date. According to the letter, operators affiliated with Alibaba conducted about 28.8 million exchanges with Claude using roughly 25,000 fraudulent accounts between April 22 and June 5, 2026.
Distillation is a training shortcut: a developer queries a stronger model at scale and uses its outputs to train a smaller, cheaper system, capturing much of the leading model’s behavior without bearing the original research cost. Anthropic flagged three earlier “industrial-scale” campaigns in February — attributed to DeepSeek, Moonshot and MiniMax — and says the Alibaba operation was substantially larger.
The complaint arrives alongside a tightening US posture on frontier AI. The administration recently directed Anthropic to suspend access to its latest models, Fable 5 and Mythos 5, under export-control measures aimed at keeping advanced capabilities out of Chinese hands. Anthropic’s letter effectively argues that the gap those controls are meant to protect is being eroded through the API itself — a harder problem to police, since distillation works through ordinary model queries rather than any breach. It also sharpens the case for usage monitoring and account-verification rules as a complement to hardware and weight restrictions.