The United States and China are in regular talks to iron out details of a framework announced last month for handling AI-related incidents and institutionalising bilateral dialogue, multiple sources told the South China Morning Post.
The technical teams of the two countries aim to make AI cooperation a potential deliverable when presidents Xi Jinping and Donald Trump meet on the sidelines of the Asia-Pacific Economic Cooperation (Apec) summit in Shenzhen next month – their third face-to-face talks this year.
Three sources said the artificial intelligence issue had taken on a more prominent role in discussions between the two sides as they explored ways to formalise cooperation in the field.
The sources spoke on condition of anonymity because they were not authorised to discuss the matter publicly.
Beijing and Washington have conducted intensive working-level discussions following Xi’s state visit to the US in September, when they announced plans to establish a bilateral channel for AI-related incidents and launch a China-US AI dialogue.
The White House fact sheet and the Chinese government’s account of the eight outcomes from last month said the leaders had agreed to set up the dialogue mechanism to exchange views on the “risks and benefits” of artificial intelligence and to open a “bilateral communication channel” for AI-related incidents, with the next meeting scheduled for November.
However, the initiatives remain largely declarative for now.
The Chinese embassy in Washington said on Thursday that Beijing “remains open to maintaining communication” with Washington through the AI dialogue mechanism.
“China puts equal emphasis on development and security in terms of AI. We take seriously the inherent and secondary risks of AI,” it added.
A source said Washington and Beijing aimed to have a basic framework for addressing AI-related crises in place by the time leaders gather in Shenzhen, with operational details to be worked out later.
“Adding flesh will come at a later date,” the source added.
Another possible outcome of the two leaders’ Shenzhen meeting could be “an announcement establishing a continuing institutional mechanism for AI cooperation, involving government officials, research institutions and industry representatives”, a second source said.
The two countries agreed to begin formal talks on AI governance following the Trump-Xi summit in Beijing in May, but the meeting was repeatedly delayed due to differences over the scope of the discussions and scheduling issues.
The first AI dialogue was eventually folded into the broader economic and trade talks between US Treasury Secretary Scott Bessent and Chinese Vice-Premier He Lifeng in New York, just days before the Xi-Trump summit in Washington. The SCMP was the first to report on the merger of the two tracks.
The November Apec meetings in Shenzhen “will provide a good opportunity to give the AI dialogue a more substantial political and institutional dimension”, one person familiar with the issue said.
AI “has emerged as one of the most politically important areas in which Washington and Beijing might demonstrate that strategic competition does not preclude reasonable cooperation”, the source added.
The source distinguished between an AI safety and incident-management agreement, which is already under discussion, and a broader AI economic cooperation framework, which would be “considerably more ambitious”.
However, another source said that while the two sides aimed to provide institutional cover to the AI dialogue framework, it would not be an “easy task” given differences in their domestic regulatory structures and the competitive nature of the US-China relationship.
The prospect of bilateral cooperation has coexisted alongside an intensifying AI race between Washington and Beijing, underscoring anxieties in the US that China is quickly eroding America’s technological edge.
Despite growing concerns over AI safety, the Trump administration has rejected calls to slow the development of its frontier AI models, pointing to competition with China.
Trump has instead placed the onus on AI companies to police themselves, meeting leading industry executives at the White House last month and backing a voluntary pact for firms to monitor and audit their own systems. He called the agreement “morally binding”.
As Chinese firms have narrowed the gap with their US counterparts in advanced AI models, American tech companies and government officials have stepped up accusations that Beijing is extracting capabilities from US models to speed up the development of its own systems.
“The technical word for stealing and copying American AI models is distillation. So, the Chinese distil our models, and they can never get ahead of us,” Bessent said last month.
Model distillation is a widely used AI training technique in which a smaller model learns from the output of larger ones.
China has dismissed the US allegations, describing distillation as “a common practice in the AI sector”.
The bilateral discussions are being held as Washington and Beijing compete to shape global AI governance, with the US seeking to build a network of trusted partners through its Pax Silica initiative, while China advances the World Artificial Intelligence Cooperation Organisation. -- SOUTH CHINA MORNING POST
