The U.S. Government is further extending the restrictions on China AI from chips and equipment to the model level. On July 21, United States Treasury Secretary Besent indicated that the United States would examine whether there were signs of intellectual property misappropriation in the open source AI model from China; if the act was found to have been committed, the United States would be able to impose sanctions on China AI.
At the Fox Business Channel on the same day, Becent stated that the United States Government supported the development of open source models, but did not accept the misappropriation of intellectual property rights, especially if the overseas model was found to be “poaching” the results of United States businesses, which would allow the United States to impose sanctions. Bloomberg then took the lead in reporting on this statement.
The White House's pressure on China continues to expand.
This statement comes against the backdrop of the continuing tightening of US policy on China's AI. Prior to that, Washington had restricted China ' s access to advanced AI chips and continued to strengthen export controls. Today, the signal from the U.S. is that future targets may no longer be limited to hardware and supply chains, or may point directly to the AI model itself.
Axios had previously reported that the Trump Government was considering a total ban on the open-source model in China, although this was also questioned in part. Although the relevant measures have not yet been formally landed, the latest statement indicates that stronger options are indeed being discussed within the United States Government.
The controversy is on the distillation of the model.
One of the core elements of the discussion around the so-called misappropriation is model distillation. The technology usually refers to the migration of some of the capabilities of large models to smaller, more deployable systems. A number of AI companies in the United States have continued to warn in recent months that foreign actors may replicate their technology and reissue it in open source form.
However, there is no consistent agreement within the industry that “stilling is a theft”. Microsoft CEO Nadra publicly questioned this statement earlier this month. He stated that there was a contradiction in the position of large model companies, which claimed their right to be based on open data training models, while imposing strict restrictions on distillation.
In the United States, AI is also exposed to copyright risks.
Intellectual property disputes are not limited to Chinese companies. In the United States, AI has been facing continuing legal pressure on its own training practices. This week, Anthropic was allowed to launch the payment process of its $1.5 billion settlement programme, after the judge found that the company had illegally downloaded and preserved millions of copyrighted books for training AI.
There are those who believe that China AI's pursuit of America cannot simply be attributed to distillation. Hugging Face CEO Clem Delangue has recently indicated that distillation is only a small factor, which is also being used by United States companies themselves. In his view, the more important reason why China is moving faster is the strength of its research team and its more open and collaborative approach to AI development.
If the U.S. finally formally applies the sanctions tool to China's AI model, the focus of US-China science and technology competition will further extend from chips to algorithms and model ecology, and the compliance and commercial pressure on the firms concerned may rise in parallel.