Policy
Geneva Is Trying To Turn AI Governance From Speeches Into Institutions
The UN's AI for Good Global Commission and Geneva governance meetings show a push to build international AI institutions before private labs and national governments harden the rules on their own.
By Michael C ·
The global AI governance debate is moving to Geneva with a familiar problem and a new urgency: everyone agrees artificial intelligence is becoming strategic infrastructure, but no one has yet built the institutions that can govern it across borders.
The United Nations and the International Telecommunication Union are using the AI for Good platform to convene technology leaders, governments and international officials. The commission arrives alongside broader Geneva discussions over whether AI governance should be led by national governments, private labs, clubs of allied states, or a more representative global institution.
That question is not procedural. It will shape who gets access to advanced models, who defines safety standards, how poorer countries participate, and whether AI rules become a geopolitical privilege rather than a global public good. The UN's strongest argument is legitimacy: AI risks do not stop at borders, and cybersecurity failures, disinformation, labor disruption, model concentration and military escalation can affect countries that did not build the underlying systems.
If AI governance is designed only by the United States, China, Europe and the largest technology companies, many countries will receive rules rather than help shape them. That could weaken adoption, trust and enforcement. Global governance cannot work if most of the world sees it as someone else's industrial policy.
The challenge is capacity. The UN can convene, but it does not automatically have access to frontier model internals, data-center economics, chip supply chains or real-time deployment evidence. A serious AI institution would need technical staff, evaluation capacity, incident channels and political independence, not just a stronger calendar of summits.
In the absence of binding global institutions, private AI labs are writing many of the norms themselves. They decide model access, red-team processes, acceptable use policies, release timing, developer terms and what counts as enough safety evidence. That is not inherently illegitimate; the labs have expertise and operate the systems. But private governance becomes fragile when revenue, competition and national-security pressure pull in different directions.
Geneva's opportunity is to build connective tissue before the rules harden elsewhere: shared evaluation practices, reporting formats, model-risk taxonomies, compute-disclosure norms and assistance for countries that lack their own AI safety institutes. A universal AI treaty is unlikely in the near term, but common standards are more plausible and may matter first.
Standards are less dramatic than treaties, but they can become powerful. Once procurement rules, insurance markets, corporate boards and regulators begin expecting a common safety dossier, companies have a reason to comply even before a global law exists.
The governance conversation also has to include development. Many countries want AI for health systems, education, agriculture, public administration and disaster response. If safety rules only restrict access without supporting capacity, the system will look like protectionism. A credible global institution would help countries evaluate tools, build local talent, protect data, negotiate with vendors and participate in standards-setting.
Geneva will not solve AI governance in a week. But it can change the shape of the problem by moving the debate from speeches about shared values toward institutions that can inspect systems, compare evidence and give more countries a meaningful voice before the rules harden.
Topics: United Nations, AI governance, Geneva, AI for Good