UN AI governance dialogue returns May 3–4, with common priorities on the agenda

The second Global Dialogue on AI Governance is scheduled for May 3–4, 2027, in New York, with co-chairs tasked with helping governments agree on shared priorities. Ahead of those talks, November 12 commentary from the Center on International Cooperation proposes a division of labor: independent scientists establish the evidence; governments decide what to do with it.

A meeting with a follow-through mandate

The permanent representatives of Saint Kitts and Nevis and the Republic of Korea will lead the second cycle. The president of the UN General Assembly appointed them co-chairs on October 16. CIC’s notice gives the meeting’s dates and city, but does not include a start time, specific venue or official watch link.

Their assignment extends beyond convening the event. They must hold intergovernmental consultations aimed at common understandings on priority areas for international AI governance. Those consultations will feed into the high-level review of the Global Digital Compact in September 2027.

The first Dialogue, held in Geneva, organized discussions into four clusters and identified broad areas of agreement. But its co-chairs’ September summary also recorded continuing differences over the scope, priorities and institutional forms of AI governance. The next cycle must work from that unfinished political foundation.

Shared facts cannot settle political choices

The Independent International Scientific Panel on AI is required to present its findings at every Dialogue. It released a preliminary report in July 2026 and has begun publishing thematic briefs. Its 40 independent experts are appointed by the General Assembly, span all regional groups, and cover technical AI, ethics and societal impacts.

Writing for CIC, Joris de Mooij argues that this work can help diplomats manage three problems: too much information, evidence that arrives after technological change, and uncertainty about which findings to trust. Smaller delegations face particular difficulties, he writes, though even well-staffed missions can lack enough technical expertise.

His proposed approach would frame each consultation around a specific governance question. Participants would distinguish established findings from uncertainties, then examine competing policy options. The Panel would supply a common factual starting point, not resolve disagreements that belong to governments.

Proposed priorities, not an adopted agenda

De Mooij recommends choosing topics through three filters: scientific evidence, member-state interest and existing initiatives that could carry cooperation forward. These are CIC recommendations for shaping the process, not a confirmed agenda for the New York meeting.

• Child safety: De Mooij identifies it as the clearest candidate. The Panel has announced a brief, and several member states launched a children’s-rights coalition in Geneva.
• Governance tools: The Panel’s preliminary report finds that dozens already exist but are fragmented and rarely measure real-world effectiveness. Further Panel work could inform discussion of how those tools work together.
• Evidence gaps: For areas including labor-market effects and the global AI supply chain, he proposes targeted written submissions and expert briefings rather than treating unsettled questions as established findings.

The proposed output is practical rather than all-or-nothing: identify cooperation that can proceed without universal agreement, record remaining disagreements, and flag areas ready for intergovernmental consideration. De Mooij also recommends allowing new scientific findings to update discussions, while sending unresolved research questions back to the Panel.