The three U.S. agencies accused DeepSeek and five other Chinese AI companies of illegal distillation

The National Security Agency, FBI, and CISA of the United States released a joint report, naming DeepSeek, Dark Side of the Moon, Alibaba, MiniMax, Leap Star, and Zhipu. The U.S. side accuses the aforementioned companies of batch calling the outputs of Claude, GPT, Gemini, and Grok to train their own models since at least the end of 2024, involving millions of requests and billions of tokens.

The report lists specific models, with DeepSeek accused of using multiple versions of Claude, GPT, Gemini, and Grok to generate training data for R1 and V3; Dark Side of the Moon is accused of using Claude Fable 5 data to train Kimi K3 and GPT-4o data to train Kimi K2. The U.S. claims these companies also bypass regional restrictions through numerous accounts, third-party API aggregators, and gray intermediaries, and attempt to extract the hidden reasoning processes of the models using prompts, labeling such behavior as malicious distillation.