Anthropic accused three Chinese AI startups—DeepSeek, Moonshot and MiniMax—of using tens of thousands of fake accounts to harvest Claude’s outputs and “distill” them into rival systems, producing more than 16 million interactions in violation of its terms. The company said the campaigns targeted high-value reasoning and agentic capabilities and warned that distilled models lack safeguards, posing national-security risks if open-sourced. Framing the activity as evidence for stricter export controls on advanced chips, Anthropic argued that limiting compute access would curb both direct training and illicit distillation. The disclosure follows OpenAI’s recent warning to U.S. lawmakers about DeepSeek’s targeting of American AI leaders; the named Chinese firms did not immediately comment.
Related articles:
– UK AI Safety Summit 2023
– NIST AI Risk Management Framework
