Anthropic accuses Alibaba of massive model distillation attack

US AI company Anthropic wrote to the US Senate, accusing Alibaba of using about 25,000 fake accounts to have 28.8 million conversations with its models from April 22 to June 5, calling it 'the largest known distillation attack.' Distillation is an AI training method that uses outputs from stronger models to build smaller ones.

An Anthropic spokesperson said fighting illegal distillation threats requires coordinated action between government and industry. However, critics point out that Anthropic has also scraped internet data for AI training. Elon Musk slammed Anthropic for massive training data theft, calling its behavior hypocritical.

In May, after Anthropic released its flagship model Claude Opus 4.8, users found that when asked about its identity, the model claimed to be Alibaba's Qwen or DeepSeek, sparking widespread discussion about model distillation and double standards.