Over two earlier pieces, we argued that the first UN Global Dialogue on AI Governance would be judged on two things: whether its process could actually include the people it claimed to speak for, and whether its agenda would go beyond AI’s visible harms to reach the questions of power and extraction that are sitting underneath. We were in Geneva for the week. This is what we saw, measured against what we said, and what the roughly ten months until the New York session should be used to fix.
Convening this Dialogue at all, in the current climate for multilateralism, was an achievement. The intent to build a genuinely multi-stakeholder process is real, and the co-chairs stewarded a difficult first outing with care. What follows is a case for making the second session fill the gaps and deliver what the first could not.
Process
In the first blog we flagged that speaking slots were being handled in ways that excluded people, with speaker lists shared at the last minute and no transparency about who was chosen. At the main event, the same pattern held. Speaker lists were expanded continuously, with names added until the last minute, which meant several people on the list were never informed they had been selected. Some learned only on the day they were due to speak that their name was on it. Others did not show up for their statements at all, because they never knew they were expected . In some cases the guiding questions for a panel changed the night before, leaving speakers working late into the night to rework what they had already prepared.
Sessions were built around sequential statements rather than exchange, and floor interventions were capped at three minutes, which made the event far less of a dialogue than its name promises. Panel questions were often either too broad or too narrowly drawn to bring out a real range of views. The format did accommodate a great many people to speak, but it allowed very little actual discussion to happen. The cluster co-chairs were not part of the panels themselves, which left their function unclear to participants.
Some of the difficulty was structural. Running the Global Dialogue, the WSIS Forum, and the AI for Good Summit in the same window split everyone’s attention. In practice it kept the human rights and policy community separate, physically and thematically, from the innovation and technical crowd. Bringing those two worlds into the same conversation is part of what a Dialogue like this exists to do, and the scheduling worked against exactly that.
Underneath all of this ran the problem the whole series has been about. Global South voices were marginal throughout, with only a handful of Global South-led side events on the programme. Declining funding across the Global South is part of the mechanism here. It shapes who can afford to travel and who can afford to organise a side event, and therefore who gets heard.
Substance
The second blog argued that the Dialogue risked governing AI’s symptoms while leaving its foundations untouched. What we saw was a slightly different version of the same problem. Most panels stayed at the level of diagnosis. They described harms, named the divides, and set out what was going wrong, but stopped short of moving toward concrete outcomes, commitments, or red lines. The framing of a “dialogue of dialogues” signalled from the start that little was expected to be negotiated, and a techno-solutionist assumption ran through much of the discussion. The preliminary report from the scientific panel could have helped push past diagnosis, since it consolidated a good deal of existing evidence into language useful for engaging governments. But it landed just days before the conference, too late to build the agenda around it.
Two absences stood out against what we had argued. The first was gender. Despite sustained, coordinated outreach ahead of the event, there was no meaningful discussion of gender in the Dialogue. Given that gendered harm runs through every layer of the extraction story, from data labour to biometric exclusion to technology-facilitated abuse, this is a striking gap. The second was the uneven treatment of human rights. A dedicated human rights cluster was a real win, but human rights were not mainstreamed across the other clusters, and the question of liability received little attention. Children’s rights were the exception, drawing real engagement, though that too risks collapsing into simplistic fixes like banning technology.
The Road to New York
The more encouraging story of the week was outside the plenary, and it points to what the second session should build on. The agenda that needs to reach New York is, in effect, the one we have been arguing for across this series, and it is now backed by a joint civil society front.
That agenda starts with treating data as something communities steward rather than something extracted from them, alongside mandatory environmental and human rights impact assessments across the AI value chain. It calls for consent to sit at the centre of how training data is gathered, so that indigenous and traditional knowledge is not absorbed without recognition or return, and so that communities keep a real say over how knowledge about them is collected and used. And it calls for binding public accountability standards across the whole value chain.
It puts gender back where it belongs, as a structural dimension of safety, with technology-facilitated gender-based violence treated as a safety failure in its own right and gender impact assessments required before high-stakes systems are deployed.
On process, the remedies are already obvious. Confirm speakers once they are placed on the list, and make any change to panel format at least two weeks out rather than the night before. Communicate clearly what co-chairs and co-facilitators are for. Provide real visa and logistical support for Global South participants, which matters enormously, because the next session is in New York, where visa barriers are expected to be considerably steeper than in Geneva. Between now and then, the global IGF sits at the midpoint of the two Dialogues and offers a ready-made, tested space for the incoming co-chairs to consult and refine the agenda.
A new set of co-chairs will take over in September, under the incoming Secretary-General, with only about ten months until the New York session. That is not long, but the diagnosis is done, the shared agenda is written, and the coalitions to carry it are organised. The first Dialogue showed the gaps. The second has enough time, and now enough direction, to close them.
