Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Claude Distillation Attacks

Anthropic has identified and disrupted large-scale unauthorized distillation attacks against its Claude model, which originated from seven China-based labs. The labs involved include those affiliated with Alibaba, Moonshot, DeepSeek, Z.ai, and MiniMax. These attacks utilized a machine learning technique known as knowledge distillation, a method that can be used legitimately for training purposes. In this context, knowledge distillation involves a powerful AI model acting as a teacher. Anthropic's action suggests that these specific instances of distillation were unauthorized and potentially malicious. The use of knowledge distillation in these attacks highlights its potential for both legitimate and illicit applications.