Story by Daragh Thomas
Anthropic accused three Chinese AI companies of running 24,000 fraudulent accounts to siphon capabilities from its Claude chatbot, in what may be the largest documented case of AI model theft to date.
DeepSeek, Moonshot AI and MiniMax generated over 16 million exchanges with Claude, violating terms of service and geographic access restrictions.
The labs used a technique called distillation, where a weaker model trains on the outputs of a stronger one, to extract Claude’s most advanced reasoning, coding and tool-use capabilities.
Anthropic’s head of threat intelligence Jacob Klein said the company has “high confidence these labs were conducting distillation attacks at scale.”
What The Labs Were After
DeepSeek ran over 150,000 exchanges specifically designed to make Claude walk through its reasoning step-by-step, generating chain-of-thought training data that could be fed directly into a competing model.
Moonshot AI hit 3.4 million exchanges targeting reasoning and coding.
The Glorious Quran and Science 
Leave a comment