US probes China AI theft claims, may impose sanctions

The United States is preparing to investigate allegations of intellectual property theft concerning artificial intelligence models, with potential sanctions against China on the table, according to Treasury Secretary Scott Bessent. Speaking on Fox Business, Bessent affirmed that the administration opposes "intellectual property theft" and will act if foreign models are found to have exploited the work of U.S. companies.

The move comes amid growing concerns in the U.S. tech sector and government about China's accelerating progress in AI. Chinese companies are rapidly developing open-weight models, which allow for external download of trained parameters, aiming to close the gap with leading U.S. firms like OpenAI and Anthropic. Moonshot AI's recent Kimi K3 model, for instance, has reportedly surpassed some U.S. competitors in industry benchmarks.

“

If we find, particularly overseas models, that they are stealing the output of our great companies, we have the ability to sanction them for that type of theft.

Bessent specifically mentioned "distillation," a technique where a smaller, weaker model is trained using the outputs of a more powerful existing model. He referenced a letter from AI firm Anthropic to the U.S. Senate Committee on Banking, Housing, and Urban Affairs, accusing Chinese tech giant Alibaba of launching what it called the "largest known distillation attack" to date. Bessent also revealed that "watermarks" of U.S. large language models have been detected on numerous Chinese models, which he deemed "unacceptable."

"We have the ability to sanction them for that type of theft," Bessent stated, indicating that the investigation would commence in the coming days or weeks. The U.S. and China are reportedly planning AI talks for September, with Bessent slated to represent the U.S. The report also noted that both Anthropic and OpenAI have faced their own accusations of data misuse, including Anthropic settling a lawsuit over unauthorized book downloads and The New York Times suing OpenAI and Microsoft for alleged copyright infringement in model training.

“

We have found watermarks of our U.S. large language models on many Chinese models, which is unacceptable. We will be investigating this in the coming days or weeks.