Anthropic Says 5 Chinese AI Labs Hit Claude With 190 Million Distillation Attacks
Updated
Updated · Business Insider · Sep 11
Anthropic Says 5 Chinese AI Labs Hit Claude With 190 Million Distillation Attacks
3 articles · Updated · Business Insider · Sep 11
Summary
Nearly 190 million exchanges hit Claude between May and July, Anthropic said, accusing Alibaba, Moonshot, DeepSeek, Zhipu and Xiaomi of harvesting its outputs to train rival models.
151 million of those exchanges came from Alibaba alone through 3,500 fraudulent accounts, while Moonshot rerouted 23 million user requests and DeepSeek 12.1 million in 14 days in July.
Anthropic said some rerouted prompts exposed sensitive Chinese and Russian government and military data, including PLA-linked CCTV footage and information tied to a Russian government database.
To curb further attacks, Anthropic said Claude now gives less detailed reasoning traces and users appearing to operate from China, Russia or Iran will face identity checks.