Not too long ago, Geneva experienced an unusual event. AI researchers, tech executives, and human rights attorneys were seated across from diplomats who had devoted their careers to discussing trade quotas and nuclear disarmament. They were all attempting to reach a consensus on what regulations, if any, should apply to systems that none of them fully comprehended. According to a number of accounts, it was similar to the early internet governance debates, but with much more anxiety in the room and higher stakes.
The 2024 Global Digital Compact gave rise to the United Nations’ Global Dialogue on AI Governance, which will take place in Geneva in July and may be the most ambitious multilateral attempt to impose some order on the development of artificial intelligence to date. It’s genuinely unclear if it will amount to anything. However, the fact that it exists at all indicates where this discussion has come from.
For years, the effort has been growing. The Bletchley Declaration from 2023 provided a common vocabulary for discussing frontier AI risk; “AI safety” became more than just a talking point in Silicon Valley. This was expanded into a network of safety institutes spanning ten nations and the European Union at the 2024 Seoul summit. Then, in February 2025, Paris occurred, and everything changed. Pledges for investments predominated. Bletchley had started a conversation about safety, but it was overshadowed by the language of adoption and economic opportunity. Earlier this year, India hosted a summit aimed at reintroducing the Global South. These events have been successful thus far in attracting political attention, but they haven’t been very successful in producing legally binding results.
The summits seem to have been ignoring the underlying issue. Tech companies, mostly based in the US and increasingly in China, are investing close to $700 billion to develop AI infrastructure without waiting for global agreement. In its 2025 Technology and Innovation Report, UNCTAD pointed out that these businesses prioritize profit over the general welfare and that, in the absence of regulation, the gap between the winners of AI and everyone else will most likely grow. Without sufficient safeguards, AI could worsen inequality and digital divides in ways that disproportionately hurt the most vulnerable nations, according to UN Secretary-General António Guterres.
One could argue that the UN is trying to create in a matter of years what other international organizations took decades to establish. The 40 top experts who make up the recently established Independent International Scientific Panel on AI are tasked with producing evidence-based reports every year. The panel is loosely modeled after the IPCC, a climate science organization whose deliberate, slow evaluations ultimately changed international climate policy. Whether or not member states consider its conclusions to be worthwhile will determine whether or not that analogy is valid.
Pursuing AI governance through the UN is not idealistic for many developing nations. It’s historical memory mixed with pragmatism. Countries that witnessed the development of the internet’s governance structures without them and then spent years attempting to catch up view the current situation as an opportunity to help shape the terms before they are set in stone. According to Tony Oweke, a research fellow at the Council on Foreign Relations who previously represented 134 developing nations in UN AI negotiations, the Dialogue serves as a safeguard against future reliance. In cities like Brasília, Jakarta, and Nairobi, where discussions about AI are increasingly focused on sovereignty rather than just safety, this framing strikes a chord.
The problem is that fragmentation has already begun. Over eighty nations have developed advanced AI strategies based on varying risk tolerances, regulatory philosophies, and definitions of what constitutes appropriate AI behavior. The deployment of AI systems across borders will likely result in inconsistent outcomes and more obvious accountability gaps than anyone is willing to acknowledge in the absence of a coordinating mechanism.
Whether the July Dialogue will result in anything more lasting than its predecessors is still up in the air. The timing is crucial: between 2027 and 2028, a formal review of the Global Digital Compact is anticipated, which could either strengthen or discreetly abandon the UN’s AI mandate. This summer’s events in Geneva will be more significant than they may appear from the outside. If this fails, AI won’t become uncontrolled. It implies that whoever moves the fastest will control AI, which is a completely different matter.
