Anthropic says Alibaba illicitly extracted Claude AI model capabilities

What We Know

Anthropic, the U.S. developer of the Claude family of AI models, has publicly accused Chinese technology giant Alibaba of running a campaign to “illicitly” extract its models’ capabilities. Multiple outlets report Anthropic described the activity as a coordinated effort to distill or reproduce Claude’s functionality and that the company sent a letter briefing U.S. officials about what it called a brazen attack on its systems.

Coverage frames the episode as a distillation or capability-extraction campaign rather than a single data breach: reporting uses terms like “distilling” and “orchestrated” and one outlet called it the largest known attack on Claude models. Related accounts note a secondary market of resellers in China offering access to Claude tokens at steep discounts — reporting referenced claims that some resellers sell tokens at roughly 70–90% below official API prices — which commentators link to the alleged extraction activity. Available reports list the allegation and the briefing to officials but provide limited technical detail about how the extraction was carried out, how Anthropic linked it to Alibaba, or what specific evidence was shared publicly.