The inaugural session of the UN’s Global Dialogue on AI Governance, held over two days in Geneva in July 2026, marked an important attempt to bring together a diverse set of stakeholders across the globe. First conceptualised as part of the Global Digital Compact of 2024 (GDC), the Dialogue has been designed to function as “an inclusive space within the UN for governments and stakeholders to deliberate on today’s most pressing AI challenges”.
Repositioning the spotlight
At a time when avenues for multinational and multistakeholder cooperation are shrinking rapidly, the Dialogue represents a rare opportunity to re-animate global cooperation in AI governance. In fact, it can be seen as a direct response to the growing power asymmetries within this landscape, both between governments and among different stakeholders – industry, civil society, and local communities.
The appointment of the permanent representatives of UN Member States such as El Salvador and Estonia — two relatively smaller countries from Central America and Northeast Europe — as Co-Chairs reflects an effort to shift the centre of gravity away from the usual clique of Western nations that has historically dominated such forums.
Over the last few months, the Co-Chairs have led several virtual, in-person, and written consultations, bringing civil society and a broader range of countries into the AI governance conversation, thus creating space for these participants to articulate their concerns and priorities directly. Most strikingly, more than half of the written suggestions came from Asia and Africa, while civil society accounted for the largest share of submissions across stakeholder groups.
The merits of dialogue in AI governance
In addition to the diversity of stakeholders invited to contribute, the Global Dialogue also offered a more open and collaborative engagement format. Participants were invited not only to share region- and sector-specific perspectives on AI governance, but also to suggest how the Dialogue itself should be structured, which existing initiatives it should build on, and what distinct value it could add. In doing so, it commendably did not presuppose its own standalone value or take its potential effectiveness for granted. Of course, it remains unclear how, and to what extent, these suggestions will be incorporated in practice. But some of the major criticisms around overlapping AI governance convenings and nominal inclusiveness were, at the very least, acknowledged upfront.
A wider catalogue of issues discussed
In terms of the substance of the Dialogue itself, there were four key thematic clusters the Jan-July conversations were structured around. These revolved around AI opportunities, bridging access and capacity divides, safety and trustworthiness in AI, and human rights.
The Dialogue began with a presentation on a preliminary report prepared by the Independent International Scientific Panel on AI, an initiative that was also born out of the GDC. Both the report and the conversations in Geneva acknowledged the increasing gap between rapidly improving AI capabilities and effective risk management methods, and expressed concern over the growing fragmentation of governance approaches across regions.
Issues that the civil society has long voiced — power concentration among a few tech conglomerates, capacity deficits of Global South countries, and AI’s adverse impacts on human rights, social systems, and the environment — emerged as clear throughlines in the report and in some of the sessions that took place on the sidelines of the Dialogue. In many ways, the Dialogue brought into a high-level intergovernmental forum conversations long confined to grassroots coalitions and critical academic forums.
The limitations of the Dialogue’s form and substance
Despite best efforts to engender participation from multiple stakeholder groups and cover a wide variety of AI governance themes, the Dialogue seems ultimately constrained by the scale of its own ambitions. There is only so much that even a genuinely multistakeholder process can achieve through dialogue alone, particularly when many of the problems under discussion are rooted in deeper geopolitical and economic structures that the Dialogue — and the UN Scientific Panel Report — are not engaging directly with. This lack of confrontation with deeper political-economic questions of AI governance is evident across several aspects of both these initiatives.
The persistence of techno-developmentalism
Most conspicuously, it was visible in the Dialogue’s — and UN’s broader digital governance paraphernalia — techno-developmentalist assumption that greater access to AI capabilities is inherently desirable, and that limited access is primarily a deficit to be overcome. Dedicated discussions in Geneva on AI opportunities and capacity-divides (and their priority ordering vis-a-vis safety and human rights themes) were emblematic of precisely this belief. While these are important topics, perhaps even more so for Global South countries, they obfuscate questions around which actors — and what kind of capital — benefit from the proliferation of a brittle and risky technology like AI. This framing also precludes any questions around the very effectiveness of AI within national or international development projects. This criticism has long been directed towards the UN’s earlier initiatives which positioned digital innovations as a critical driver for human development.
Missing agents, missing accountability
Another related limitation was the limited attention paid, across both the Dialogue and the Scientific Panel’s report, to identifying the actors and institutional choices driving many of the harms under discussion. Although both acknowledged concerns such as the concentration of AI development, corporate responsibility and the potential for government misuse, these issues were seldom examined in ways that clearly located responsibility or enabled meaningful accountability. This includes the role of frontier AI companies in developing increasingly powerful and potentially harmful systems, as well as that of governments promoting AI adoption despite its social and environmental costs or deploying these technologies in harmful ways. Topics such as AI use in international warfare or domestic surveillance were largely missing from the Dialogue.
Unclear pathways to action
A third concern around the Dialogue’s ability to make headway in global AI governance stems primarily from its very nature. Ultimately, the Global Dialogue remains exactly that: a dialogue. While the UN is capable of bringing stakeholders together and legitimising certain anxieties of civil society, it lacks the explicit institutional or legal authority required to convert these dialogues into enforceable commitments or hold powerful actors to account.
It is particularly difficult to overlook this limitation at a time when many — if not all — of the AI harms being discussed have already been thoroughly documented and discussed in forums that pre-date the Dialogue. Over the years, researchers, civil society organisations, affected communities, and international institutions have consistently generated evidence, raised concerns, and suggested potential solutions. The main challenge is no longer about merely opening up more avenues for conversation, but rather translating this extensive body of research and discussion into tangible action.
In many ways, then, the current design of the Global Dialogue as a primarily discursive platform falls short of the needs of the less powerful actors it purports to serve, particularly if their participation is not matched by the institutional will and courage needed from the Dialogue — or any other global AI governance effort — to hold technocapitalist elites to account.
The absence of the United States is another facet of the Geneva event that cannot be ignored. Given the outsized role the US has historically played within the UN verse, as well as its influence over the global AI industry and the policy choices of other governments, its decision to remain outside of or peripheral to the Dialogue sends an important political signal. This does not mean that the Dialogue should reproduce a US- and UK-centred model of multilateralism – indeed, part of its value may lie in creating greater space for countries and perspectives that have long been marginalised within global governance. At the same time, meaningful multilateralism cannot entirely sidestep the participation of a state whose policies often shape the positions and behaviour of smaller or closely aligned countries.
May 2027 and Beyond
Many of the issues discussed above are not unique to the Dialogue. They are a product of the global moment we find ourselves in: one in which capital, capacity, and decision-making power within the AI realm remain concentrated in the hands of a few. And where well-intentioned avenues for consultation have not delivered the global cooperation urgently needed in a post-generative-AI world.
But this is precisely why the Dialogue must step up to the plate. Rather than replicating familiar models of multistakeholder consultation, it should use its convening power, its institutions’ legitimacy, and norm-setting role to experiment with a bolder model for AI governance. It must look towards new ways not just to conduct dialogues but also to orchestrate action in a deadlocked governance landscape.
A few shifts could help the Dialogue begin moving in this direction:
One of the main objectives of Dialogue is to ensure that AI governance reflects the priorities of all nations, not just the most technologically advanced. But power hierarchies exist not only at the level of national governments but also within nations. Bringing government officials, industry executives, and civil society into the same room does not automatically flatten deeply entrenched hierarchies between them.
Civil society or non-state actor participation therefore needs to be more firmly institutionalised within the Dialogue, with community voices and bottom-up evidence carrying as much legitimacy as high-level scientific and governmental processes. This means creating formal pathways through which evidence generated by civil society and affected communities can shape the Dialogue’s priorities, findings, and eventual recommendations, rather than remaining an input into a process whose terms are largely set by high-level state officials.
The UN cannot continue limiting itself to a relatively vanilla version of AI governance centred on broad principles and voluntary cooperation. The Dialogue should clearly distinguish between issues that still require further deliberation and harms that have already been extensively documented, and use the latter to build a clearer set of expectations from frontier AI companies and governments — including mandatory disclosures, accountability requirements, and enforceable red lines.
But enforcement cannot depend only on whether powerful governments or companies choose to cooperate in good faith. The Dialogue should also help equip governments, regulators, and civil society organisations with practical tools to demand greater accountability — whether through model disclosure requirements, procurement standards, incident-reporting mechanisms, evidence-gathering frameworks, or other legal and institutional avenues. In this sense, its role should not stop at articulating norms, but extend to making those norms easier for different actors to actually use and enforce.
If the Dialogue is serious about repositioning the Global South, it should move towards a more distributed model of engagement across regions. This could take the form of a hub-and-spoke structure, with regional processes that are not simply one-off consultations feeding into a central UN convening, but recurring spaces with the mandate and resources to shape the Dialogue’s priorities over time.
This would also allow the UN to draw on evidence that emerges much closer to where AI systems are actually being deployed and experienced – strengthening the Dialogue’s own legitimacy and sustainability as a mediation platform for AI governance.
If the Dialogue does not address some of these fault lines now, it risks becoming a missed opportunity at precisely the moment when such spaces are becoming harder to sustain.