Later this week, Presidents Trump and Xi will sit down in Beijing. AI is on the agenda. Just two weeks ago, the three of us also traveled to China with a mission to discuss AI. We came not as government representatives, but as researchers and policy practitioners. We made the American case on one of the most consequential risks of our time: AI loss of control. While there, we listened carefully to how counterparts on the other side of the Pacific are approaching the same problems, and assessed whether there is a basis for the kind of structured dialogue that serves both nations’ interests. (In short, we learned there is.)
The trip was a whirlwind. We arrived in Shanghai on April 22 and departed Beijing only six days later. While still on the ground, a Wall Street Journal profile of Treasury Secretary Bessent described a standing effort within the Administration to monitor AI risks ranging from runaway systems to catastrophic misuse, and confirmed that AI will be on this week’s summit agenda. And on April 24, South China Morning Post reported that Senator Daines would lead a bipartisan congressional delegation to these same two cities to help lay the groundwork. Taken together, these sent a clear signal that engaging China on AI, including on shared risks, serves American interests. We agree.
Our shared focus for the trip was AI loss of control, or the risk that advanced AI systems become too autonomous and too capable for humans to reliably oversee and correct. We each contributed something different to the conversation. Mark brought a philosophical framework, drawing on both Western and Confucian traditions, to articulate why this moment demands restraint and stewardship. Karson brought a technical perspective on how AI systems can be governed from within, using layered monitoring approaches, interpretability research, and AI-assisted oversight to manage systems that are growing too complex for traditional human supervision alone. And Steve brought a national security framework developed by the team at IST: a loss of control Indications & Warning methodology that applies decades of intelligence community practice to the problem of detecting early warning signals before they become crises.
Together, those three lenses — philosophical, technical, operational — made for a more complete argument than any of us could have made alone.
We presented at the Shanghai Forum at Fudan University and at the 12th China and Globalization Forum in Beijing. But the formal sessions were only part of the picture. In the hallways, over meals, and in a series of side meetings and private discussions, we had the chance to engage with a variety of researchers, officials, industry representatives, and others for some of the most substantive conversations of the trip.
That is precisely why American researchers and policy practitioners need to be in these rooms. Staying out of these conversations does not make the risk smaller. It only means American voices are absent when the frameworks are being built.
We have no illusions about the complexity of the U.S.-China relationship, or the real differences in values and systems that exist between our countries. The case we made in both cities was simple: AI presents risks neither country can manage alone, and common cause on those risks is not a concession. It is a strategic choice.
The quiet preparatory work we did in Shanghai and Beijing — testing frameworks, building relationships, identifying where common ground might exist — is exactly the kind of foundation that serious dialogue requires.
Following this week’s summit, our work towards common cause on AI loss of control will continue. July’s World AI Conference in Shanghai will be the next opportunity to go deeper.
Steven M. Kelly is Chief Trust Officer at the Institute for Security and Technology. He is a retired FBI special agent and served twice on the NSC staff, most recently as Special Assistant to the President for Cybersecurity and Emerging Technology.
Mark Beall is President of the AI Policy Network. He served as the inaugural Director of AI Strategy and Policy at the Pentagon’s Joint Artificial Intelligence Center and has testified before Congress on AI and national security.
Karson Elmgren is Senior Researcher on the international strategy team at the Institute for AI Policy and Strategy, where he focuses on China. He previously worked at RAND, the Center for Security and Emerging Technology, and OpenAI.
