The UN Opens Its First Global Dialogue on AI Rules

On 6 and 7 July, delegates from governments around the world gather at the Palexpo convention centre in Geneva for the first United Nations Global Dialogue on AI Governance. It is the closest thing the world has yet had to a single table where every country, rather than a handful of powerful ones, is invited to argue about how artificial intelligence should be governed. Whether that table can agree on anything is a separate question.

The dialogue grew out of the Global Digital Compact that member states adopted in 2024, and it runs alongside the AI for Good summit and the annual World Summit on the Information Society meetings. The centerpiece is the debut of a new body: the Independent International Scientific Panel on Artificial Intelligence, a group of 40 experts drawn from every region and co-chaired by the computer scientist Yoshua Bengio.

The science arrives first

The panel released a preliminary assessment ahead of the meeting, and its central finding is blunt. "AI capabilities are outpacing both scientific understanding and governments' ability to adapt," the report warns. Policymakers are caught in a bind the panel lays out plainly. They need solid evidence to write good rules, yet the evidence keeps falling behind a technology that changes faster than anyone can measure it.

UN Secretary-General Antonio Guterres welcomed the assessment, framing it as the moment the field finally has an agreed scientific baseline to work from. "The science is here," he said. A fuller report is due next year. The panel is modeled loosely on the Intergovernmental Panel on Climate Change, the idea being that governments are more likely to act together when they share one trusted account of the facts rather than a dozen competing ones.

A crowded field of rulebooks

The dialogue lands in the middle of a busy season for AI regulation, and much of it is pulling in different directions. Washington is racing to finish a voluntary framework for frontier models, which we covered in a look at its August deadline. Brussels, meanwhile, has just delayed the toughest parts of its own AI Act. One capital leans on light-touch pledges, another on binding law, and the UN is trying to find a thread that runs through both without the authority to compel anyone.

That is the honest limit of what Geneva can do. The Global Dialogue produces conversation, shared reports, and side agreements, not enforceable treaties. Critics reasonably ask what a voluntary forum achieves when the companies building the most capable systems, and the two governments most invested in them, can simply proceed at their own pace.

The counterargument is that the alternative is worse. Climate diplomacy was slow, frustrating, and full of unmet pledges, and it still built the shared vocabulary that later action depended on. If AI governance is heading somewhere similar, this week in Geneva is the part where everyone at least starts reading from the same page.