Anthropic accused Moonshot AI of covertly routing thousands of user requests to Claude and passing the responses off as its own output. In the same threat intelligence report, Anthropic said seven China-based labs ran campaigns to copy Claude’s capabilities and that the largest one, attributed to Alibaba, topped 151 million exchanges between May and July.

Anthropic said the report also detailed operations it disrupted involving potential biological weapons research and a Russian-aligned espionage campaign against more than 20 organizations. It also described an Iran-linked effort targeting Israeli and Jewish diaspora individuals and compiling handbooks on U.S. naval positions, and a Yemen-based cell’s use of Claude to write guidance, navigation and control software for a guided rocket, a ballistic missile with a 2,000+ km range goal and a hypersonic glide vehicle variant. Anthropic said it disrupted every operation in the report and shared some findings with authorities and other AI companies.