What stayed with me from the recent UN General Assembly (UNGA) discussions on artificial intelligence (AI) is that for the first time, more than 60% of world leaders — 122 countries — recognized AI’s importance to society as a whole, but just 20 countries and the European Union signed on to a joint declaration pushing for a new global body to run safety tests and enforce common standards.
That gap between recognizing AI’s potential and acknowledging that its risks must be addressed collectively is the critical divide where the speed of technological innovation still outpaces global consensus, and shared principles still need to be translated into mutual accountability.
At last week’s 81st UNGA, AI leaders warned the Security Council that increasingly capable systems could pose risks no country can manage alone. They called for greater international cooperation, particularly on testing and reporting serious incidents.
The need for cooperation is clear. What it should look like is less so.
Governments have different views about the role and reach of global oversight and what AI governance and regulatory frameworks should actually look like. National security, economic competition and different political traditions all shape those views. Countries may agree that certain risks deserve attention, yet disagree about who should set standards, what information should be shared, whether oversight should be voluntary or mandatory, and what, if any, enforcement mechanism should be applied.
Those differences are real, and they will affect whether any international approach is workable and trusted. A comprehensive global framework may be difficult to achieve.
But we shouldn’t let the search for one perfect model prevent progress on practical measures. Governments have worked together on shared risks in other areas, even when they disagreed about authority or enforcement; the Chemical Weapons Convention and Chemical Safety that brought together 193 countries, the Nuclear Non-Proliferation Treaty and even the formation of the U.N. itself have demonstrated that international cooperation can be achieved.
We can do the same with AI. Nation-states can improve how they share information about serious incidents, coordinate on evaluating systemic risks, and support independent scientific expertise. Common terminology and technical methods would also make it easier to compare evaluations and understand what incident reports are telling us.
These steps wouldn’t mandate what every country must permit or prohibit. But they could help governments see emerging risks sooner, assess the evidence together and make better-informed decisions.
We're already seeing the first instances of international cooperation with regard to AI, including the UN Independent International Scientific Panel on AI; UN GA Resolution A/RES/79/325 (Global Dialogue on AI Governance); the Global Digital Compact; and UNESCO's Global Forum on the Ethics of AI (its underlying framework represents the first global standard-setting instrument on AI ethics).
Now we need to make progress. Here’s the reality: AI systems are changing quickly, and the information needed to assess them isn’t always readily available. Independent research and credible testing that help distinguish demonstrated capabilities from speculation can better equip policymakers to make well-grounded decisions on warranted safeguards, striking a balance between capturing critical risks and preventing overly restrictive measures.
We need a shared understanding of how to advance the safe, secure, and responsible development of AI. National action is essential, but global coordination is indispensable.
I also keep coming back to who actually has a seat at the table. Many countries will be affected by AI without having the resources to develop it or shape its direction; they should still play a meaningful role in deciding how it’s deployed and what safeguards apply, as well as access to the expertise and institutional support needed to participate. International frameworks will be more legitimate and more durable if they reflect the perspectives of the people and communities affected.
There’s a practical lesson here for the private sector as well. Expectations around safety, transparency and accountability are evolving across jurisdictions. Companies should be ready to explain how their systems work, where they may affect people or important processes, and how risks and incidents are identified and addressed. That takes clear ownership, appropriate testing and monitoring, and coordination among product, engineering, legal, privacy, security and policy teams from the start and throughout the product lifecycle.
As UN Secretary General Antonio Gueterres appropriately stated in his General Assembly address, "We need a shared understanding of how to advance the safe, secure, and responsible development of AI. National action is essential, but global coordination is indispensable.”
The UN meetings didn't solve how AI should be governed. However, global cooperation can still advance now with or without a treaty. Sharing what we know about incidents, aligning on evaluating risks, and bringing more countries to the table to have a real voice. This level of information sharing can begin now. That’s not a treaty, but at the very least a good start.
