Anthropic has accused three Chinese AI firms—DeepSeek, Moonshot AI, and MiniMax—of illicitly extracting capabilities from its Claude models. The company claims these labs used approximately 24,000 fraudulent accounts. These accounts generated over 16 million interactions to harvest data in violation of service terms and regional restrictions.
The campaigns utilized a technique called distillation to transfer knowledge from Claude to newer Chinese models. These sophisticated operations targeted advanced reasoning and coding functions from Anthropic’s most capable systems.
Anthropic warned that these actions allow foreign labs to replicate restricted capabilities. This process potentially undermines U.S. export controls designed to limit China's progress in artificial intelligence.