Washington Accuses Six Chinese Groups of AI Model Distillation

American security agencies are targeting six Chinese companies that they accuse of building their AI systems on the distillation of American models. There is probable knowledge by Beijing and potential effects on military capabilities and cyberattacks. Beijing rejects these allegations and emphasizes the country's scientific autonomy.

American agencies highlight military effects and probable knowledge by Beijing

According to U.S. security authorities, distillation campaigns are being conducted on a large scale, using methods deemed aggressive, malicious, and specifically targeted. They believe these methods would allow China to reduce development costs while enhancing its military and cyberattack capabilities, which could be deployed against the United States and its partners. These agencies also indicate that the Chinese government would likely have been aware of these actions.

Six named groups and a method presented as central

The NSA, FBI, and Cybersecurity and Infrastructure Security Agency assert that distillation forms the foundation of model development at several Chinese companies. According to the agencies, the scale and sophistication of the campaigns show that this approach is not merely supplementary. Six players are cited, including Alibaba, Moonshot, and DeepSeek, and they are accused of having used models from Alphabet (Google), Anthropic, OpenAI, and SpaceX.

Beijing contests and recent context revives the accusation

Chinese authorities refute these allegations. Mao Ning, spokesperson for the Ministry of Foreign Affairs, claims that the performance of AI models developed in China is based on advanced scientific and technological independence. Since the emergence of Chinese artificial intelligences, the United States has maintained that China reproduces American models through distillation. This assertion has been renewed recently, particularly after Moonshot's Kimi K3 impressed the public.