Anthropic has accused Chinese AI companies Alibaba, Moonshot and DeepSeek of conducting unauthorized, large-scale distillation of its Claude models by routing queries through networks of intermediate platforms. The company said the firms relayed millions of their own users’ requests to Claude, including queries containing sensitive information such as location data and passwords, without users’ knowledge. Anthropic said Alibaba generated more than 151 million Claude exchanges between May and July, including a peak of nearly 3 million exchanges a day across more than 3,500 fraudulent accounts. It said Alibaba used the outputs to train its Qwen models and for reinforcement learning and model-architecture research. Anthropic attributed more than 23 million exchanges between May and July to Moonshot, including nearly 300,000 requests allegedly relayed to Claude during a 10-day period. It also reported more than 12 million suspected DeepSeek distillation attacks over 14 days in July. The company said the activity raised privacy and terms-of-service concerns, and that U.S. officials have likened the technique to industrial-scale theft. Futurum Group CEO Daniel Newman said the allegations challenged portrayals of Chinese open-weight models as independently developed. Alibaba, Moonshot, DeepSeek and Xiaomi did not immediately respond to a request for comment. Further company responses, benchmark results, public leaderboard changes and actions by Anthropic CEO Dario Amodei remain key developments to watch.