illicit distillation
-
Chinese AI Labs Secretly Leveraged Millions of Claude Exchanges for Model Training
Anthropic reports that Chinese entities, including Alibaba, Moonshot, and DeepSeek, engaged in unauthorized AI model training. They used “illicit distillation,” training their models on outputs from Anthropic’s Claude without permission. This practice infringes intellectual property and may violate data privacy, as sensitive information was accessed. Alibaba’s campaign involved millions of interactions, while Moonshot and DeepSeek also systematically diverted user requests to Claude for training data.
