DeepSeek and Alibaba face US accusations of copying advanced AI models
American intelligence agencies said that six major Chinese artificial intelligence companies systematically used advanced US models to train their own systems. According to Washington, through millions of requests, Chinese developers obtained billions of tokens from Claude, GPT, Gemini and Grok.
This is stated in a joint warning from the US Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency (NSA) and the Federal Bureau of Investigation (FBI).
The American agencies named six companies: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. According to US intelligence, since at least late 2024 they have been conducting large-scale so-called knowledge distillation campaigns.
This refers to a method in which a less powerful model is trained on the responses of a more advanced system. Distillation technology itself is widely used in the artificial intelligence industry and is a legitimate way to train models.
However, American agencies claim that Chinese companies used it aggressively and on an industrial scale, trying to systematically obtain closed capabilities and features of advanced American models.
According to CISA, Chinese developers received billions of tokens through millions of interactions with various versions of Claude from Anthropic, GPT from OpenAI, Gemini from Google and Grok from xAI.
American intelligence agencies believe that such activities could have been carried out with the knowledge of the Chinese authorities. The document also states that large-scale distillation campaigns have become one of the key tools for the development of the Chinese AI industry.
In the opinion of American agencies, this approach allows Chinese companies to significantly reduce costs for research, computing power, electricity and development of fundamental technologies necessary to create advanced models from scratch.
The US also said that the identified methods violate the terms of use of American AI services. According to intelligence, Chinese companies distributed activity among various model providers, cloud platforms and infrastructure services to complicate detection of the campaigns.
CISA, NSA and FBI called on American AI companies to strengthen control over suspicious accounts and anomalous requests, track sharp spikes in the use of new accounts and exchange information about potential distillation campaigns.
Companies were also recommended to change model responses in cases where the system detects signs of mass automated data collection for training a competing AI.
As noted by NBC News, this became one of the most detailed public accusations by the US government against specific Chinese artificial intelligence developers. The Chinese Embassy in Washington did not immediately respond to a request for comment from the outlet.
