měi guó美国zhǐ指liù jiā六家zhōng guó中国rén gōng zhì néng人工智能gōng sī公司dà liàng大量xué xí学习měi guó美国mó xíng模型
Three U.S. government agencies recently said that six Chinese artificial intelligence companies, including DeepSeek, Moonshot AI, and Alibaba, may have obtained a large number of answers from advanced U.S. models to train their own models.
This method is called “model distillation.” It lets a smaller model learn from the answers of a larger model, reducing training time and costs.
The method itself is common and is not necessarily illegal. However, the U.S. side believes that these companies used fake accounts, network relay services, and other methods to bypass usage restrictions and made millions of requests.
The U.S. side is concerned that this could affect the technology and competitiveness of U.S. companies, and that it could also be used for military purposes or cyberattacks.
China’s Ministry of Foreign Affairs denied these claims, saying that China’s AI development relies on its own scientific and technological capabilities, and called on the United States to stop making false accusations.
U.S. agencies have not released complete records, so it is currently difficult for outside observers to independently verify these allegations.