The first Global Dialogue on AI Governance: Positions and Recommendations Moving Forward

The inaugural Global Dialogue on AI Governance took place in Geneva on 6–7 July 2026. As UN Tech Envoy Amandeep Singh Gill emphasized, the Dialogue cannot be a one-off; it has to iterate toward action. In that spirit, this commentary provides both a look backward and a look forward.

The opening speech by the Secretary-General was cognizant of the challenge ahead

The UN Secretary-General delivered an impassioned speech at the Global Dialogue. He repeated his call for the AI Environmental Transparency Initiative, the AI Child Safety Pledge, and a ban on lethal autonomous weapons. The military domain is outside of the mandate of the Global Dialogue. However, as the SG argued: “AI does not respect that line. The same models and chips have moved into the battlefield.”

His speech contained memorable quotes like “We cannot vibe-code the future of humanity”. And the warning: “We may be the last generation able to set the terms on which humanity and machines coexist. The door is still open. But it will not stay open long.” One can only hope his successor will pick up where he left off.

Superpowers with different postures

China’s Minister of Industry and Information Technology, Li Lecheng, framed AI governance as “a shared task of all countries,” emphasized the equal right of all countries to develop and benefit from AI, and pointed to China’s AI Capacity Building Action Plan and its Global AI Governance Action Plan. Open source ran through the Chinese positioning as both offer and argument: Chinese open-weight models such as DeepSeek and Qwen were presented as shared assets that lower barriers, support digital sovereignty, and let users keep data local — with Tsinghua’s Xue Lan citing adoption figures for Chinese open models and warning against cross-border restrictions on an open global AI ecosystem, a formulation as much about market access and export controls as about model weights. The various Chinese statements remained consistent with Beijing’s Global Governance Initiative white paper released in June, in support of a leading UN role in AI governance.

The United States engaged, but at a distance. Its official remarks were delivered at staff level, several rungs below the roughly sixty ministerial delegations. It was also the only P5 member absent from the June interactive dialogue with the Scientific Panel. The substance was courteous and concrete (child safety, intellectual property, censorship, workers) while restating the administration’s theory of AI governance: industry as a core partner, voluntary cooperation as the only approach agile enough to keep pace, and the diffusion of the American AI stack to partner countries as a matter for bilateral diplomacy. The sharper edge appeared just before Geneva, in a State Department Substack criticizing UN capacity-building language and warning that “digital sovereignty” interpreted as national AI tech stacks would result in “synchronized mediocrity”.

In summary, the UN is a venue for the two AI superpowers to talk to the other 190+ countries. Whereas China likes the multilateral venue, the US political will for AI governance with universal scope remains limited. From the topics mentioned by the US, child safety and workers are the concerns shared across geopolitical fault lines.

The demand for responsible diffusion from the rest of the world

The G77 stressed the need to close the AI divide. AI capacity building should be a “permanent pillar” of international cooperation and the UN Global Fund on AI should move forward. They also argued that the Global Dialogue should grow “beyond exchanging views” and “connect discussion with implementation”. As the Minister of Communication and Information Technology of Saudi Arabia asked: “90% of the compute today, as well as IPs in terms of AI, are held in 2 countries. Only 2 countries. Is that the future that the United Nations accepts for more than 190 member states?” As the UAE Minister argued, history shows that we need some level of separation of power, decentralization, and an independent judiciary to avoid tyranny.

The need for diffusion also comes with a need for risk management. Especially, as the Qatari minister warned: “Dangers of insecure models do not know borders.” Overall, the statements show an emerging consensus that AI models are both high opportunity and high risk. The AI booster backlash of 2025, when some were triggered by the mere mention of “AI safety”, is over. The risks are undeniably real, as are the opportunities. As the Romanian minister emphasized, “we must be pro-innovation and responsible at the same time. These are not competing instincts, but the same duty seen from two sides.” Or as Jaan Tallinn summarized: “Most countries, including many advanced economies, lack the technical expertise to assess the most capable AI or have a real say in governing it. And because these risks are actually cross-border, even if you don’t have a say, you’re still going to be subject to the risks. The countries left out are exposed just like everyone else. So bridging the AI divide and governing the risks are not two separate agendas.”

The key question that remains is how to operationalize responsible diffusion. There is also broad agreement that, as the EU emphasized: “governance should be informed by facts and evidence”. Still, as always, the devil is in the details. Germany argued that we may need new approaches: “Society and governments cannot move forward with the policy recipes of the past. We will need new answers, for example, for the labor market of the future”. France specifically highlighted that “we must collectively tackle the issue of independent evaluation of AI models before their rollout.” Austria went even further and argued: “the world needs a binding framework for the use of AI.”

Either way, as the German minister observed: “We must shape AI, because otherwise AI will shape us”. As the UAE minister underlined, we are at a turning point, and decision-makers must live up to the task: “The decisions that we take today are not just decisions for ourselves and our generation, but for all generations that are to come.”

In summary, most countries are betting on AI benefits but remain concerned about the potential for extreme power concentration and about AI capabilities that come with increasingly serious misuse and autonomy risks. 

Three recommendations 

The second UN Global Dialogue will take place in New York, back-to-back with the STI Forum (5-6 May 2027). Below are three recommendations to ensure it reaches common understandings on AI governance that can be recorded in a UNGA resolution:

1. Appoint the new co-chairs early

The inaugural co-chairs were appointed by the President of the General Assembly roughly eight months before Geneva. The incoming PGA, Khalilur Rahman from Bangladesh, should match that timeline, naming the next pair by late autumn 2026 at the latest. Notably, the next co-chairs will not only have to produce a co-chairs’ summary but also consultations on a resolution on AI governance. To advance this, should we encourage the PGA to give the new co-chairs an explicit encouragement to produce a zero-draft of common understandings on the mandated thematic pillars based on the first Dialogue and Panel report before delegates land in New York.

2. Drive child safety to a signable outcome for a coalition

One of the clearest coalitions of action that has formed around the Global Dialogue is the Coalition for Children’s Rights and Protection in the Age of AI. This multistakeholder coalition is driven by Spain and France with UNICEF, OHCHR, and ITU, supported by the European Commission, and has been joined at launch by seventeen states from Brazil and Indonesia to Japan, Kenya, and Korea. The Preliminary Report of the Independent International Scientific Panel on AI has flagged the issue of child safety and AI as well, explicitly listing it as a topic for a potential thematic brief.

To maximize its impact on child safety, it’s advisable for the coalition to take the responsible diffusion of open AI into consideration for this challenge (e.g., watermarking downloadable models at distribution platforms for provenance in case of incidents). Child safety also matters because it is one of the few topics on which buy-in across geopolitical lines looks plausible.

3. Strengthen focused discussion formats

The many input speeches from political leaders at the first Global Dialogue offer a good stocktaking of different stakeholders’ positions. Rather than repeating this 1:1, we recommend that the Second Global Dialogue hone in on specific topics of interest through focused, iterative multistakeholder engagement. 

We thank the co-chairs H.E. Egriselda López and H.E. Rein Tammsaar, as well as their teams, for their tireless work on this process and look forward to engaging with their successors for the Second Global Dialogue.