GENEVA – For two days in the Swiss congress city, the governments of each United Nations member state are being asked to agree on something the technology industry cannot solve on its own: who is responsible when an artificial intelligence system causes serious damage, and how the world finds out when it does.
The United Nations General Assembly’s first global dialogue on artificial intelligence governance opened Monday in Geneva, attracting delegations from all 193 member states to confront a problem that the technology’s architects admit they haven’t solved. Diplomatic ambition is great. And so is the obstacle at the heart of this meeting: the computing infrastructure that makes advanced AI possible is concentrated almost entirely in a single country whose government arrived in Geneva having already rejected the central premise of the meeting.
An independent scientific committee set up by the General Assembly issued its preliminary findings at the opening of the summit, and the conclusions were deliberately stark. The 40-member International Scientific Committee on Artificial Intelligence, chaired by Turing Award winner Yoshua Bengio and Nobel laureate Maria Ressa, concluded that science currently cannot guarantee that as AI capabilities continue to increase, these systems will not cause catastrophic harm. Risks range from independent misconduct to intentional misuse by malicious actors. “Science currently cannot guarantee that AI will not cause catastrophic harm, either on its own or because of malicious users,” Bengio told delegates.
This result, translated from the diplomatic record, has a clear meaning for the 193 delegations gathered in Geneva. There is no technical guarantee that advanced AI systems will reliably follow instructions given to them. The summit is, in part, an attempt to build governance scaffolds around a category of technology whose makers cannot yet promise to stay within those scaffolds.
Negotiators are working through three interlocking proposals: a cross-border incident reporting mechanism that would require countries and companies to share information when an AI system causes significant harm; A set of international safety assessment standards against which parametric models can be measured; The liability framework for autonomous AI systems, meaning who pays when something goes wrong, is designed to operate without direct human supervision. None of this would be considered a simple negotiation under the best of circumstances. These procedures become considerably more difficult when the entity required to implement them has already indicated that it does not intend to do so.
The United States explicitly rejected what its delegation described as “centralized global governance” of artificial intelligence before the summit opened. The position is not without logic: Washington’s concern is that a UN-administered framework would give authoritarian governments meaningful oversight over AI systems that they currently cannot access on their own terms. What makes the situation awkward is the calculation. According to figures presented at the summit, US-based companies host nearly 75 percent of the world’s top 500 companies in terms of AI supercomputing capacity. The countries advancing to the Palais des Nations on Monday govern 193 members of the international system. They do not collectively rule those 75 percent.
This gap between the location of AI power and the source of demand for rules is precisely the kind of structural asymmetry that UN multilateral processes exist to address, and precisely the kind they most often fail to resolve. And Washington has been following its own path: Earlier Monday, the White House and leading AI labs moved closer to a voluntary framework for border models, a domestic approach that implicitly goes beyond the binding international accountability mechanism that Geneva is trying to build.
Ressa, the Filipino journalist who won a Nobel Prize for documenting the consequences of manipulating information, came to Geneva with a specific warning about where the current governance vacuum is actually causing the costs. She said that AI-generated content spreads extraordinarily quickly when combined with emotional manipulation, and the infrastructure needed to disseminate it on a large scale exists today, running on the same computing clusters whose concentration in US data centers made the Geneva talks structurally complex from the beginning.
The summit’s organizers, co-chaired by Ambassadors Rene Tamsar of Estonia and Igreselda Lopez of El Salvador, organized the two days around the premise that has guided UN technology diplomacy since the first multilateral action in communications: that global engagement is more difficult and more sustainable than agreement among the willing few. The presence of all 193 member states is unusual for a technology gathering of this kind. The General Assembly’s AI Governance Committee, announced just days ago with Nvidia’s Jensen Huang and Amazon’s Andy Jassy joining its roster, has made clear how difficult it is to keep both the industry and member states aligned on a single governance structure. The statement signed by 193 governments is a different document from the one signed by the major AI powers and their closest allies. Whether it’s a better tool depends on what it actually requires.
This question remains clearly on the table as the summit enters its second and final day. There has been no public indication that the White House is willing to participate in any binding outcome. Scientific Committee The full report was issued in conjunction with the opening of the summitIt has not yet been fully absorbed by the negotiators working through the text of the ruling in parallel. The liability framework, which requires legal systems in 193 countries to recognize a category of harm that most of those systems were not designed to address, is the provision with the fewest precedents and the most resistance.
What the Geneva Dialogue actually accomplished is narrower but not trivial: it produced the first scientific consensus at UN level that the systems currently deployed on a global scale carry no technical guarantee of human control. This conclusion will continue even after any statement issued on Tuesday. This is the foundation on which we will have to build all subsequent governance arguments, whether or not the state that built most of these systems agrees to be part of building what comes next.
