Home / Technology / US Warns: Chinese Firms Stealing AI Models
US Warns: Chinese Firms Stealing AI Models
9 Sep
Summary
- US agencies accuse Chinese firms of large-scale AI model data extraction.
- Companies like DeepSeek and Moonshot AI allegedly distilled US models.
- Mitigation strategies for US firms were detailed in a joint advisory.
United States intelligence agencies have issued a stern warning regarding Chinese AI companies, accusing them of "distillation activities at an industrial scale." The NSA, CISA, and FBI jointly advised that several Chinese firms have been extracting proprietary capabilities from advanced American AI models. This alleged extraction, occurring since 2024, involves obtaining "billions of tokens" through millions of requests to train their own models.
Specifically named in the advisory are DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. These companies are accused of leveraging data and functionalities from prominent US models, including OpenAI's GPT series, Anthropic's Claude, Google's Gemini, and xAI's Grok. For instance, DeepSeek reportedly used data from various Claude and Gemini models to train its own reasoning model, R1. Moonshot AI allegedly distilled data from Anthropic's Fable and GPT-4o to develop its advanced Kimi K3 and Kimi K2 models.
This warning echoes long-standing allegations by American companies regarding Chinese firms copying their technologies. OpenAI previously banned accounts suspected of such "distillation," with DeepSeek being a focus of their investigation. Anthropic has also recently accused DeepSeek, Moonshot, and MiniMax of similar large-scale campaigns. The advisory includes crucial mitigation strategies for US firms to combat these alleged malicious distillation efforts, highlighting an ongoing tension in the global AI development race.
