News Summary

Anthropic, a U.S. AI startup, has accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of conducting ‘industrial-scale campaigns’ to extract capabilities from its Claude chatbot. This practice, known as distillation, involves using responses from more advanced models to enhance less capable ones. Anthropic claims that these firms created over 24,000 fake accounts to initiate approximately 16 million exchanges with Claude, using these interactions to improve their competing AI systems. The accusations highlight concerns about model security, intellectual property theft, and the implications of AI distillation in global AI competition. Anthropic has expressed concerns about the potential military and intelligence applications of these extracted capabilities, emphasizing the need for improved security measures to prevent such attacks in the future.