Anthropic Accuses DeepSeek, MiniMax, Moonshot of Ripping Off Claude
Anthropic claims three Chinese AI firms hammered Claude with 16M+ prompts to distill its intelligence into their own models.
Anthropic is pointing fingers at three AI companies — DeepSeek, MiniMax, and Moonshot — alleging they violated its terms of service in a big way. The accusation: all three used Claude as a training crutch, firing off a combined 16 million-plus prompts to extract knowledge through a technique called distillation.
Distillation essentially means using a powerful model's outputs to teach a smaller, cheaper one. It's the AI equivalent of copying someone's homework and calling it your own.
The allegations echo what OpenAI has already told House lawmakers — that DeepSeek specifically used distillation to juice its own models. Now Anthropic is piling on with its own receipts.
The pattern is clear: major U.S. AI labs are increasingly accusing Chinese competitors of systematically siphoning their models' capabilities rather than building from scratch.