Life-or-Death AI Decisions Must Remain Human-Led, Says WAIC 2026 Panel

At the 2026 World Artificial Intelligence Conference, panelists including Xue Lan (director of Tsinghua University's Institute for AI Governance), Nicholas B. Dirks (president and CEO of the New York Academy of Sciences), and Mark Nitzberg (executive director, UC Berkeley's Center for Human-Compatible Artificial Intelligence) discussed AI agent governance. According to the panel, AI is shifting from "assisting judgment" to "autonomous action," requiring expanded accountability chains across developers, deployers, and regulators.

The panelists reached consensus: decisions with life-or-death consequences, scenarios that cannot be corrected once wrong, and all matters involving ethical and value judgments must never be led by AI. Humans may authorize AI action but cannot delegate responsibility. Each authorization should be revocable, each action traceable, and the pace of granting agency must never exceed the pace of verifying AI capability.