Anthropic Accuses Chinese AI Firms of Mass Data Theft
Anthropic accused three Chinese AI companies—DeepSeek, Moonshot AI, and MiniMax—of using more than 24,000 fake accounts to send over 16 million prompts to Claude and harvest its outputs to train their own models. The practice, known as model distillation, could undermine U.S. competitive advantage in AI and allow adversaries to access unguarded capabilities for surveillance and cyberattacks.
What happened
Anthropic accused DeepSeek, Moonshot AI, and MiniMax of creating more than 24,000 fraudulent accounts to access Claude and train their own models. The three companies allegedly sent more than 16 million prompts to Claude to siphon off output for their own products, using a technique known as distillation to train smaller student models on the outputs of Claude.
Reported impact
Anthropic stated that widespread distillation by China-based models that do not charge subscription fees could make it harder for U.S. providers such as OpenAI and Anthropic to monetize products they have spent billions of dollars to build. The company warned that illicitly distilled models may lack safeguards to prevent misuse for biological weapons development, cyberattacks, and that foreign labs distilling American models could feed unprotected capabilities into military, intelligence, and surveillance systems.
Index analysis
Anthropic said the scale of the alleged distillation activities varied, with DeepSeek generating about 150,000 interactions with Claude, while Moonshot and MiniMax logged more than 3.4 million and 13 million interactions, respectively. Anthropic stated that the campaigns are growing in intensity and sophistication.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.