US artificial intelligence company Anthropic has accused Chinese companies DeepSeek, Moonshot AI and MiniMax of stealing data from the Claude chatbot. The companies allegedly created at least 24,000 fake accounts to train their models and sent more than 16 million queries in total. 150 million of the queries went to DeepSeek, 3.4 million to Moonshot and 13 million to MiniMax.
The issue is related to a method called “distillation”, in which a larger model is used to train a smaller model. Anthropic says this approach is only legal if the company uses its own model. OpenAI and Microsoft have previously made similar allegations against DeepSeek. DeepSeek has said that its V3 model was trained on open sources and e-books.
