WAIC panel says AI must not lead decisions involving life, irreversible harm, or moral judgment
At the 2026 World Artificial Intelligence Conference in Shanghai, a panel on AI agent governance converged on a sharp boundary: AI may be authorized to act, but it must not be allowed to bear responsibility or take the lead in decisions involving life and death, irreversible consequences, or moral and value judgments. The discussion brought together Xue Lan, dean of Tsinghua University’s Institute for AI International Governance, Nicholas B. Dirks, president and CEO of the New York Academy of Sciences, and Mark Nitzberg, executive director of the Center for Human-Compatible AI at the University of California, Berkeley. The speakers said AI is moving beyond assistive judgment toward autonomous action, turning an old principal-agent problem into a digital-age governance challenge. Unlike human agents, AI systems may be misaligned with user intent, operate as black boxes, and cannot themselves carry legal responsibility. The panel argued that accountability must instead sit across the execution chain, including developers, deployers, and regulators. They also outlined three engineering requirements for trustworthy AI: strong foundations, transparent operation, and controllability during use, while calling for shared safety standards, mutually recognized testing systems, incident data sharing, and early warning mechanisms at the global level.