Anthropic accuses Chinese AI firms of illicitly training models on Claude
American company Anthropic accused three Chinese startups—DeepSeek, Moonshot AI, and MiniMax—of illegally using the Claude service to obtain data and improve their AI models. According to Anthropic, these firms made over 16 million requests using 24,000 fake accounts, violating service terms and regional access restrictions.
Source Reuters
Anthropic reported that the Chinese startups employed a technique called “distillation,” whereby a less powerful model is trained based on the outputs of a more powerful one. The company argues that such activities underscore the need for export controls on American chips, limiting both direct training and model distillation.
This case highlights the growing importance of regulating technology exports and AI security, signaling potential tightening of restrictions for foreign companies in the near future.
