Global Guardrails for Agentic AI: Can the UN, or Anybody Else, Provide Them? | Carnegie Endowment for International Peace

Dramatic advances in agentic AI models are rapidly outpacing governance responses, raising profound safety concerns and even questions about the future of civilization. As these capabilities explode, the world needs guardrails to minimize AI risks while allowing humanity to reap its rewards. But who, precisely, is best placed to shape the contours of such a regime?

The United Nations aims to lead the AI governance agenda, but its aspirations collide headlong with two realities. First, the United States and China, the two main AI players, are locked in a fierce rivalry to dominate this frontier technology. Second, most AI capabilities rest in the hands of private companies locked in zero-sum competition of their own. Given these dynamics, what can the UN realistically hope to accomplish in this arena?

As the world convenes in New York for the opening of the 81^STsession of the UN General Assembly, please join Carnegie’s Stewart Patrick in conversation with Lucia Velasco, co-director of the Oxford AI Diplomacy Lab, and Scott Singer, co-director of Carnegie’s China AI Initiative. Our speakers will identify priorities for the governance of agentic AI and suggest potential pathways for establishing needed guardrails. The discussion will consider the relative roles of the UN, major and middle powers, subnational actors, and tech companies themselves in any future governance arrangements.