UN Global Dialogue Co-Leads Issue Call to Action on Safe, Secure, and Trustworthy AI

This past July, 170 countries came together in Geneva for the first UN Global Dialogue on AI Governance. Today, the Co-Leads of Cluster 3 on Safe, Secure, and Trustworthy AI — one from government, one from civil society — are laying out five steps to carry that momentum forward.

Paula Bogantes Zamora, Minister of Science, Innovation, Technology and Telecommunications of Costa Rica, and PAI’s CEO, Rebecca Finlay, have written jointly to the Dialogue’s Co-Chairs ahead of UNGA.

They propose the following steps before the Dialogue reconvenes in New York in 2027.

Align the evidence base with the deliberation, by mapping the Independent International Scientific Panel’s working groups against the Dialogue’s clusters.

Draft a common reference baseline, bringing existing commitments from international law, the Hiroshima Process, the EU AI Act, and the AI Summit pledges into a single point of reference that regions can adapt.

Institutionalize continuity between Dialogues, through a standing intersessional workstream that tracks implementation over time.

Fund participation so that Global Majority inclusion is substantive rather than nominal, with the Secretariat budgeting explicitly for it.

Pilot interoperability, through cross-border regulatory sandboxes under joint supervision, with results reported back in 2027.

Geneva showed growing consensus across stakeholder groups on what we need to achieve. The next test is whether we can turn that consensus into action through a shared baseline and interoperability tested in practice.

Read the letter below.

H.E. Ms. Egriselda López Permanent Representative of El Salvador to the United Nations Co-Chair, Global Dialogue on AI Governance

H.E. Mr. Rein Tammsaar Permanent Representative of Estonia to the United Nations Co-Chair, Global Dialogue on AI Governance

We write jointly, as Co-Leads of Cluster 3 on Safe, Secure, and Trustworthy AI, to express our gratitude for your stewardship of this inaugural Dialogue, and to offer reflections drawn from our session, including interventions from participants across government, industry, and civil society, together with written recommendations submitted during the discussion.

It is a tremendous achievement to convene 170 countries, alongside industry, civil society, and academia, around a technology that is moving this quickly. Under your leadership, the Dialogue has created a truly global forum in which AI governance can be deliberated by all who will live with its consequences. Our responsibility now is to ensure that what began in Geneva arrives in New York with momentum.

As Co-Leads, we were tasked with convening this cluster to build convergence on the infrastructure needed for global AI safety and interoperability, including identifying common baselines for frontier systems, common methods to verify them, and the shared resolve to hold systems to those standards.

That mandate resolved, in practice, into a single question — one the cluster’s interventions kept returning to from every angle: how do shared evidence and direction become governance that actually connects across borders? Participants converged, with unusual consistency, on six priorities:

A stronger, independent scientific evidence base, with the Independent International Scientific Panel on AI connected to other independent scientific panels and AI Safety Institutes, and resourced to reflect the diversity of languages and contexts it serves.

Accountability built on shared assurance: disclosure across the AI value chain, expanded independent evaluation, and interoperable benchmarks, standards, and incident-reporting mechanisms so that one jurisdiction’s lessons have the potential to prevent another’s harms.

Interoperable governance built on existing foundations — an inclusive shared baseline of good practice, common taxonomies, and practical tools such as joint evaluations and regulatory sandboxes — rather than a single regulatory model.

Governance shaped by all regions as equal partners, and backed by the resources, including funding and technical support, that equal participation requires.

Oversight that evolves with the technology, including clearly assigned, rights-protecting responsibility as AI systems gain autonomy.

Continuity between Dialogues, so recommendations are tracked, implemented, and built upon rather than rediscussed at each convening.

We believe five steps are within reach before the Dialogue reconvenes in New York in May 2027, and we respectfully submit them for your consideration, including through reference in the Co-Chairs’ report, which we see as the natural vehicle for turning convergence into a mandate for the intersessional period.

Align the Independent International Scientific Panel on AI and the Dialogue. Convene a working session between the Panel and the cluster co-leads to map the Panel’s seven working groups against the Dialogue’s four clusters, reporting back well ahead of 2027.

Draft a common reference baseline. Commission a small drafting group of the co-leads, with Global Majority and civil society representation from the outset, to map existing commitments (international law, the Hiroshima Process, the EU AI Act, and the pledges of the AI Summits, among others) into a single reference baseline of good practice, adaptable to national and regional context, for your consideration ahead of the 2027 Dialogue.

Institutionalize continuity between Dialogues. Establish a standing intersessional workstream, reporting to the Co-Chairs, to track implementation of this Dialogue’s recommendations. As one contribution, Partnership on AI (PAI) offers its in-development Global AI Progress Hub, a public record of progress on responsible AI practices, including the AI Summit Commitments, as a possible input to this work, without any expectation of endorsement.

Fund participation to support Global Majority inclusion. Request the Secretariat to budget explicitly for travel and participation to sustain engagement by lower-capacity governments, researchers, and civil society, and to report on the engagement achieved.

Pilot interoperability. Invite interested member states, or work through partner organizations such as the OECD, to pilot one or two cross-border regulatory sandboxes under joint supervision, with results shared back to the Dialogue in 2027. The Government of Costa Rica and PAI would be honored to co-convene this work as a continuation of the cluster.

The convergence our cluster witnessed in Geneva suggests the field is ready for its next step toward governance that is measured by evidence, built by all regions as equal partners, and capable of evolving as the technology does. Each Dialogue gives us a rare chance to measure where the field actually stands against a shared baseline, and to be honest about what is still missing.

Closing the gap between innovation and accountability starts by asking different questions than the ones currently driving the field: not how fast a system can be deployed, but who it serves and who answers when it fails; not how powerful a system is, but how verifiable its safety is, and against what evidence.

We hope the New York Dialogue, building on Geneva, will mark real progress on these questions.

We are grateful for your leadership, and we stand ready to support you, the Independent International Scientific Panel on AI, and our fellow cluster leads in carrying this work forward.

Please accept, Excellencies, the assurances of our highest consideration.