The Global Dialogue on AI Governance is unlike any international forum on AI governance that has come before it. Established by the UN General Assembly through Resolution 79/325, it is the only recurring process of its kind with universal membership with a genuine mandate to speak for all countries rather than for just those with the loudest voices or the deepest pockets. That makes it worth engaging with. It also makes the process failures visible so far all the more consequential.
A Bigger Table with Higher Stakes
Previous international conversations on AI governance, such as the AI Safety Summit in Bletchley Park, the Paris AI Action Summit, the India AI Impact Summit, were hosted by individual governments with their own priorities and limited membership. Important conversations happened in each of those spaces, but they were ultimately shaped by the host country’s agenda and drew on a narrow slice of the world’s governments and communities. The UN Dialogue is structurally different. Every country has a seat. That is not a small thing in a governance landscape increasingly fragmented between competing national AI strategies and a handful of dominant corporations.
The Dialogue is designed differently. It is anchored in the General Assembly, draws on the seven thematic areas in Resolution 79/325, and organizes its work around four clusters: AI opportunities and implications; bridging AI divides; safe and trustworthy AI; and human rights, transparency and accountability. These clusters were developed through an intergovernmental process led by the co-chairs, informed by a multi-stakeholder consultation process that drew over a thousand submissions globally.
Taken together, these are the right ingredients. A universal mandate and a rich body of civil society input create the conditions for a genuinely representative outcome. But a strong mandate is only as good as the process that delivers on it. And on process, the Dialogue has already accumulated a set of concerns that deserve to be named clearly.
Emerging Concerns
The online consultations held in the lead-up to the July forum have laid bare a tension that runs through the Dialogue’s design.
The numbers look impressive. Virtual consultations have drawn hundreds of registered speakers, and over a thousand written submissions were received in response to the Global Dialogue’s call. But volume and the quality of participation are two different things.
In the first consultation, held on 18 March 2026, over 600 participants registered for a session lasting three hours, with 3-minute speaking slots. That is mathematically enough time for roughly 60 speakers. Barely ten percent of those registered got to say anything at all.
By the follow-up consultation on 12 May 2026 several participants who had registered, shown up, and were prepared to speak did not appear on the speakers’ list. When participants asked how the list had been compiled, they were told priority had been given to those who had not spoken in earlier sessions. But in practice, people who had spoken before appeared on the list, while others who had also been on previous lists but had not managed to get a speaking slot were left off. The criteria, in other words, were neither transparent nor consistently applied.
There is another, less visible problem. Speaker lists are not published in advance, but are dropped into the chat at the start of the call. Participants arrive without knowing whether they or their partners will be called to speak. They cannot strategize in advance. Should they speak on behalf of their own organization or take on a coalition role representing partners who are unlikely to be called? Without the list, they cannot adjust their intervention accordingly and decisions become guesswork. With most participants never reaching the microphone, speaking time is scarce and hard-won. When the list appears only as the session begins, that time cannot be used well.
Then there is the question of notice. Registration windows for consultations have opened and closed within three to four days. For organisations with dedicated staff and established monitoring systems, that is tight but workable. For the civil society groups from the Global Majority, which are often small organisations with under-resourced teams and people working across time zones and in contexts where institutional support is thin, it is not enough. Yet these are precisely the communities this Dialogue most needs to hear from. A process that announces participation opportunities and closes them within days is not designed for broad inclusion, whatever its stated intentions.
There is also a deeper issue that no amount of improved speaker selection can fully resolve. Being allowed into the room for 3 minutes is not the same as having influence over what gets decided. Participation and agenda-shaping are different things. Even well-run speaking slots do not automatically translate into influence over the Co-Chairs’ summary, over who gets to help lead the thematic discussions, or over what the process frames as settled versus contested. These are the mechanisms through which the Dialogue will actually shape global AI governance, and civil society’s role in them remains unclear.
This pattern is not new. Those who participated in the Global Digital Compact (GDC) consultation process will recognise a similar pattern: consultations had not always been designed for global participation in terms of format and timing, draft versions and schedules went unpublished, leaving stakeholders unclear about how their input had contributed to successive drafts. The concern was that opacity around basic procedural information made meaningful participation harder than it needed to be. The risk is that it becomes normalised, that the procedural opacity of consultation design gets treated as an unfortunate logistical reality rather than a governance choice that can and should be changed.
Ensuring Meaningful Participation
Beyond the consultation format, three further procedural issues deserve attention before July.
The co-chairs have invited expressions of interest from Member States and relevant stakeholders to co-chair one of the four thematic discussions. The letter sets out some criteria, such as geographic balance, diversity of perspectives, relevant expertise, level of representation, but these are broad enough to give the co-chairs near-total discretion. There is no indication of how the criteria will be weighted, what a strong application looks like, or how the final decisions will be communicated. Organizations are being asked to put themselves forward for a significant role without knowing in any meaningful sense how they will be assessed.
For side events, each organization is limited to one application. In principle this is reasonable, given the constraints on space and time. But in practice, it falls unevenly. Larger, better-resourced organizations can distribute their presence through affiliated entities and partners with institutional capacity. Smaller organizations, particularly those from the Global Majority with limited networks and no affiliated bodies to coordinate with, are working with a single shot. The problem is not the limit, but the way it interacts with existing inequalities that the process has not acknowledged or tried to address.
Finally, no interpretation services will be provided for official side events. For a Dialogue that presents itself as genuinely global and inclusive, this is a significant gap. Interpretation is a basic condition for meaningful participation by civil society from non-English-speaking contexts and should not be seen as a luxury. Its absence effectively restricts full participation to those who are fluent in the languages in which side events happen to be conducted. For civil society organisations working in languages other than the dominant ones, it is the difference between being able to participate fully and being present in name only.
What Needs to Change Before Geneva
None of these problems are insurmountable. The July forum has not happened yet. There is still time to change some of what is in motion, and the second Dialogue in New York in 2027 offers a fuller opportunity to build a better process from the ground up.
Transparent speaker selection criteria should be published in advance of every consultation, taking into account regional diversity, gender balance, and meaningful representation of Global Majority civil society. Participants who are not selected deserve to know why. NETMundial+10 navigated similar logistical challenges with greater transparency and set a useful precedent.
Speaker lists should be shared in advance, at minimum a few days before the call, so that participants can plan how to use the time available. Three minutes used strategically, in coordination with others, is worth far more than three minutes prepared in the dark.
Prioritising coalition submissions has a clear logic: one intervention representing a coalition of organizations working across different regions and communities can surface common ground more efficiently than a dozen separate statements. The trade-off is positions that don’t survive negotiation get dropped, and the voices of smaller community-specific organizations get lost in the push for a unified message. The right design makes room for both the coalitions and the community representatives to speak to what is theirs alone.
The Co-Chairs’ summary, to be produced following the July forum, will be the primary record of what the Dialogue produced. That record needs to reflect where there was genuine agreement and where there was not. Minority views, dissenting positions, and the perspectives of actors who are structurally less powerful in these spaces deserve to appear in it. The hundreds of written submissions that civil society organizations invested time in preparing must be shown to have shaped the outcomes, not merely been received.
The Dialogue’s universal membership is its most distinctive asset. That asset means nothing if the communities most affected by AI systems leave Geneva with no more influence over AI governance than they had before they arrived.
TGI will be in Geneva for the Dialogue. We will be watching, participating, and reporting back on what we see.
