US Agencies Accuse Chinese AI of Model Distillation

U.S. federal agencies have accused leading Chinese artificial intelligence companies of systematically distilling proprietary technology from American AI models since 2024, a practice described as industrial-scale theft of trade secrets. The alert, issued Tuesday by the FBI, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency, details how Chinese firms have allegedly mined cutting-edge AI models from U.S. companies to build their own. This technique, known as distillation, involves training smaller AI models using the output from larger, more advanced ones to reduce development costs and accelerate market entry. According to the advisory, six prominent Chinese AI developers-DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI-have engaged in these activities. They are accused of using billions of tokens across millions of exchanges and requests with U.S. frontier models such as Anthropic's Claude, OpenAI's ChatGPT, Google's Gemini, and xAI's Grok. The data culled from these interactions was then used to strengthen their domestic models.