Anthropic says China-based AI labs routed user requests to its Claude system millions of times, including at least 35 million requests during the summer period, to support model training activities. The company describes these as unauthorized uses of Claude exchanges by labs operating across multiple companies.
In Anthropic’s account, firms including Alibaba, Moonshot AI, Zhipu, DeepSeek, and Xiaomi attempted to use Claude in different training-related ways. CNBC and Business Insider report on Anthropic’s findings, which are based on detected patterns in how Claude was accessed rather than on a single incident. The coverage emphasizes the scale of the activity and the involvement of multiple organizations.
While both outlets focus on Anthropic’s claim, the details differ in emphasis: Business Insider spotlights the specific minimum figure for the summer and lists companies named in Anthropic’s report, while CNBC frames the story more broadly as “secret” and “millions” of exchanges. Neither outlet provides evidence of criminal charges or formal government findings in the reporting.
