U.S. agencies accuse Chinese AI companies of systematically stealing U.S. model technology

(Deutsche Welle Chinese website) The U.S. Cybersecurity and Infrastructure Security Agency and other agencies said on Tuesday (September 8) that China's top artificial intelligence companies, including DeepSeek and Moonshot AI (Dark Side of the Moon), have systematically stolen technology from American models with the goal of competing for technological hegemony.

Agence France-Presse predicts that the above warning was announced on the eve of Xi Jinping’s visit to the United States and his summit with President Trump. Therefore, it is likely that artificial intelligence will become an important topic in the Trump-Xi meeting.

For some time, the U.S. government and U.S. AI companies have repeatedly accused Chinese companies of stealing artificial intelligence technology through a process called "distillation."

The Cybersecurity and Infrastructure Security Agency is part of the U.S. Department of Homeland. "Chinese AI companies are systematically extracting functions and capabilities that are unique to U.S. AI company models," reads the notice released on Tuesday. They "use industrial-scale knowledge distillation operations that form the core of their AI development strategies, not just ancillary means," the notice noted.

“Distillation” refers to training an AI model to imitate the output of another larger and more capable model. This method is widely used in the industry, but Washington says it is used on a large scale by Chinese companies to illegally copy U.S. systems.

The notice stated that Chinese companies adopting this strategy have "significantly shortened the AI ​​development cycle and reduced the capital investment required to train cutting-edge models."

The notice also details how companies such as DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI allegedly used millions of interactions and requests to steal the capabilities of models from U.S. companies such as OpenAI, Anthropic, Google and xAI; the document stated that these actions were "likely carried out with the knowledge of the Chinese government" and had begun in the second half of 2024 at the latest.

As China's AI models become increasingly successful, Trump faces growing pressure. Companies, including many U.S. companies, favor the more cost-effective Chinese models over the more powerful but more expensive U.S. versions of Anthropic or OpenAI.

In July this year, dozens of start-ups wrote to the Trump administration, urging them to review the situation when considering banning or severely restricting foreign AI models. They suggested that if restrictions are necessary, then they should be limited to government departments. Also in July, Beijing hit back, accusing American AI of using Chinese samples to train its own models.