US and China Establish AI Safety Dialogue as Washington Seeks to Balance Cooperation With...

The United States and China have agreed to establish a dialogue mechanism to manage risks from advanced artificial intelligence, even as both governments continue competing for dominance in the technology, US Treasury Secretary Scott Bessent has said.

In prepared remarks to the National Committee on US-China Relations in New York on Tuesday, Bessent said the two countries had created a new channel focused on what he described as "super intelligence," or SI, to address shared risks and prevent misunderstandings between their respective systems.

The arrangement will allow Washington and Beijing to exchange best practices for testing advanced AI models before their release, maintain human oversight over the technology, and alert each other to serious incidents involving highly capable systems.

Bessent said the initiative was intended to manage risks without assuming that the two countries would agree on the broader direction of AI development.

"The United States and China should harbor no illusions about our differences," he said, adding that both governments would need to reconcile technological competition with their responsibilities as global leaders.

The agreement marks an effort to establish communication safeguards around powerful AI systems, but it does not amount to a commitment by either country to slow development or impose common restrictions on frontier models.

The dialogue comes as the Trump administration faces pressure to address the potential dangers of advanced AI while maintaining the United States' competitive position against China.

Washington has largely resisted broad restrictions on AI development, concerned that tighter domestic regulation could give Chinese companies an advantage in a technology increasingly viewed as central to economic and national security.

President Donald Trump has recently indicated some openness to safeguards. In a CBS interview released on Sunday, he said he did not object to guardrails but opposed regulations that could leave the United States in "a deep second position" behind China, which he said had no comparable restrictions.

Bessent's remarks follow the same policy tension. The proposed mechanism focuses on information exchange, model testing and crisis communication rather than binding limits on research, computing capacity or the release of more capable systems.

Cooperation on safety does not necessarily reduce the competitive pressures driving investment in AI. Both governments can share information about serious incidents while continuing to expand their domestic AI industries, protect sensitive technology and restrict access to strategic resources.

The United States has already imposed export controls on its most advanced semiconductor technology destined for China. It has also accused Chinese AI developers of copying American models through what it describes as industrial-scale distillation, a process in which one model's outputs are used to train another.

Washington's restrictions on access to advanced chips remain a central part of its effort to limit China's ability to develop powerful AI systems. Beijing, meanwhile, has rejected calls to slow frontier AI development.

China's foreign ministry dismissed such proposals in September, arguing that fearmongering and confrontation would disrupt global AI governance. Huiyao Wang, founder of the Center for China and Globalization, a think tank closely aligned with Beijing's thinking, said last month that the two countries should confront AI risks and manage them together.

Impact of Xi-Trump Meeting

The AI arrangement also builds on an understanding reached during Chinese President Xi Jinping's September visit, when Washington and Beijing agreed to establish an AI dialogue and a communications channel to help defuse serious incidents.

The next test of that understanding could come during the expected meeting between Trump and Xi next week on the sidelines of the Asia-Pacific Economic Cooperation Leaders' Meeting, scheduled for November 18 to 19.

Bessent is expected to join the US delegation.

The meeting will take place against a wider backdrop of tensions over trade, technology and national security. AI cooperation offers both governments an area where limited coordination could serve their interests without requiring either side to abandon its competitive objectives.

Bessent said the two countries would have to demonstrate how they could manage that balance as AI capabilities advance.

"Our interests will not always coincide, and our approach to every challenge likewise will not always converge," he said.

China's ambassador to the United States, Xie Feng, also addressed the gathering, describing cooperation between the two countries as indispensable to tackling international challenges, including artificial intelligence governance.

"Cooperation may not resolve every challenge," Xie said, "but none can be resolved without it."

The remarks suggest that both governments recognize the need for communication, although the practical value of the new mechanism will depend on how much information each side is willing to share, how incidents are defined and whether either government acts on warnings from the other.

The event took place as Nvidia CEO Jensen Huang faces scrutiny over his reported relationship with a Chinese businesswoman suspected by US intelligence officials of having links to Chinese espionage.

Huang has acknowledged being unfaithful to his wife, while Nvidia has said its board has no reason to believe that US national security or company secrets were compromised. The company's board is conducting an internal review, and Huang has pledged to cooperate.

The controversy has added a sensitive corporate dimension to Washington's relationship with Beijing, particularly given Nvidia's central role in supplying advanced processors used to develop and operate AI systems.

Huang has been an important interlocutor between the technology industry and the Trump administration, advocating for access to the Chinese market while Washington debates restrictions on advanced chip exports. He joined Trump's delegation to Beijing in May and attended a state dinner in Washington in September.

His participation in the upcoming summit is now uncertain. Other business executives, including Tesla and SpaceX CEO Elon Musk and Mastercard CEO Michael Miebach, could join Trump's delegation, according to the South China Morning Post.

The timing places the proposed AI safety dialogue alongside a controversy involving one of the industry's most influential executives and the risks of foreign intelligence activity. However, the two issues remain distinct: the dialogue concerns government coordination over advanced AI systems, while the allegations involving Huang are subject to a separate corporate review and investigation.