AI risks belong in US-China policy, but Campbell’s interview is not consensus

That distinction should not sideline AI. The relevant question is whether risks that could affect both countries warrant routine discussion, even while the governments compete. The evidence supports putting that question on the agenda while distinguishing possible harms from confirmed incidents and measured probabilities.

Shared risks warrant discussion, not alarm

The Associated Press has reported experts discussing possible shared concerns that include AI-enabled cyberattacks, biological misuse, major model failures and loss of human control. These are risks and potential topics for dialogue, not evidence that a particular incident has occurred or that its likelihood or scale has been established.

AP reported that Sun Chenghao, a senior fellow at Tsinghua University’s Center for International Security and Strategy, saw room for cooperation and cautioned against treating safety discussions as nothing more than geopolitical maneuvering. Samm Sacks, a senior fellow at Johns Hopkins SAIS, argued that leaders need not reach broad agreement before creating political room to acknowledge shared risks. Scott Singer, co-director of Carnegie’s China AI Initiative, described AI dialogue as vulnerable to shifts in the wider US-China relationship.

The practical point is modest: governments need not resolve their strategic competition to communicate about a serious incident. Such communication would not prove the risks are equally understood or eliminate them. It could, however, give officials a channel to exchange information when a crisis makes uncertainty more dangerous.

There is a proposal to examine, but not an agreement to celebrate. AP reported that Treasury Secretary Scott Bessent said the United States had proposed an AI-incident notification mechanism in discussions with China. The report does not establish that China accepted the proposal, that a mechanism is operating or what events it would cover. Those unanswered questions make the proposal a starting point for scrutiny, not evidence that cooperation is already in place.

The Quad can complement bilateral talks

The Quad provides a regional venue for practical cooperation, but its stated technology work should not be mistaken for a shared AI-risk policy. Its foreign ministers, from Australia, India, Japan and the United States, met in New Delhi on May 26, 2026. Their joint statement included critical and emerging technologies, including AI, among areas for cooperation with regional partners.

The group’s technology agenda also includes applications beyond risk management. A Japanese foreign ministry fact sheet says a Quad initiative announced more than $6 million in awards for six research projects, including agricultural work involving AI, robotics and sensing. That is evidence of regional technology collaboration—not an adopted safety framework. The Quad can help members and partners pursue practical projects, but the available material does not show it replacing direct US-China communication.

US policy also illustrates why competition and risk management belong in the same conversation. A June 5, 2026, White House memorandum directs the national-security enterprise to accelerate AI adoption and says adopted systems should be reliable, robust, steerable and controllable. Those are administration objectives, not independent evidence that deployed systems meet them. The push to adopt AI and the difficulty of building trust with China are both part of the policy problem; focusing on only one obscures the trade-off.

Make incident communication routine

The strongest argument against deeper cooperation is practical: governments competing over national security may hesitate to share information, and the US directive shows pressure to adopt AI faster. Our recommendation is to explore a narrowly defined notification protocol, with negotiators specifying reportable events, contacts and safeguards for sensitive information. Whether either government would accept such terms, or whether such a protocol would help, is unknown; the reported proposal’s design is not established.

We should support regular US-China communication about serious AI incidents while using Quad discussions for complementary regional work. Quad members have a basis for technology cooperation, not grounds to claim they share a safety framework. Any incident mechanism should be judged by whether both governments agree to it, use it and find it useful—not by the fact that it was proposed or that a former senior official discussed AI risks. That is a more credible standard for policy in a field where competition is real and some risks may cross borders.