The United States and China have agreed to hold government-level talks on AI before their leaders meet at the White House in September.
Reuters reported Tuesday that the AI talks — agreed upon at the May summit between the two countries — could take place before Chinese President Xi Jinping's U.S. visit on Sept. 24. Multiple sources said Treasury Secretary Scott Bessent will lead the American delegation.
Preparations remain at an early stage. Participants, venue and agenda are still under review, and a date has not been finalized. There is broad agreement, however, that the AI talks should precede the summit to allow sufficient discussion of the agenda.
The White House, the Department of State and China's National Development and Reform Commission declined to comment.
Bessent said in a Fox Business interview Tuesday that he would examine allegations that Chinese AI models were developed by "distilling" American ones, raising intellectual property concerns. "We are finding the 'watermarks' of American large language models in many Chinese models. This is unacceptable," he said. "We will be looking into this in the coming days or weeks."
Bessent added that if Chinese AI companies had indeed distilled American models, it could be considered "IP theft." "We have the ability to sanction them for such theft," he said.
Both countries are weighing how to respond to the threat that the other's frontier AI models could pose to their industries or national security. The United States already restricts exports of advanced AI chips to China and is also reportedly considering limiting domestic companies' use of Chinese AI models following the release of Kimi K3, a low-cost, high-performance model from Chinese startup Moonshot.
China, meanwhile, is examining ways to restrict foreign access to its advanced domestic AI models, citing concerns that Anthropic's Mithos model could be used for hacking.
kate01@heraldcorp.com