ANTHROPIC warned that three Chinese AI labs have used a technique called model distillation against Claude, generating millions of queries to copy the model. According to Anthropic, DeepSeek, Moonshot and MiniMax have generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts, in violation of terms of service and regional access restrictions.
Model distillation is described as a legitimate AI training method, but the blogs say it can be used maliciously to rapidly obtain advanced capabilities from other labs. Anthropic said the campaigns employed fraudulent accounts and proxy services to access Claude at scale while evading detection, with distinct prompts suggesting deliberate capability extraction rather than legitimate use.
The firm also noted that illicitly distilled models could be used for harmful purposes, including potential security risks such as enabling offensive cyber operations and disinformation, and that Anthropic does not offer commercial access to Claude in China or to subsidiaries of Chinese companies outside the country for security reasons.