Thematic discussion 1: AI opportunities and implications - Global Dialogue on Artificial Intelligence Governance - Day 1

Excellencies, distinguished delegates, ladies and gentlemen, allow me to convey a very warm welcome to the Thematic Cluster 1 of this first Global Dialogue on AI Governance. Artificial intelligence is transforming economies and societies at unprecedented speed. New models, new capabilities, and new applications are emerging not in years but in months. This reality poses a challenge to us all. No government, no companies, and no international organizations can match this speed of change alone. Looking ahead, we must work with all actors shaping these technologies across the international landscape. We need to follow development closely, understand their real-world impacts, identify what works, and help codify and disseminate good practices, whether they emerge from the public sector, the private sector, or public-private partnership. In this fast-moving environment, Multilateralism is not optional. It is essential to ensure that technological progress is guided by shared principles, broad participation, and common benefit. Our shared goal must be to prevent a widening AI divide, a divide in access to technology, skills, data, and computing power, but also in access to the opportunities that AI can create. This is where UNIDO brings particular value. As the unique United Nations Special Agency for Industrial Development, UNIDO stands at the intersection of technology, industry, and development. Through our work on industrial AI, including the Global Alliance on AI for Industry and Manufacturing, AIM Global in short, and our global network of Centers of Excellency, we help translate the promise of AI into practical solutions that create jobs, strengthen industries, and build local capabilities. Distinguished delegates, the decisions we make today will shape how AI serves humanity for decades to come. This dialogue must therefore be more than an exchange of views. It must be a catalyst for action to bridge divides, build capabilities, strengthen cooperation, and turn principles into practical outcomes. Let us leave this room with a shared commitment to work across sectors, across regions, and across disciplines to ensure that AI becomes a force for inclusive and sustainable development. The opportunity is before us, but the window to act is limited, so we must act fast. With that, let me hand over to my colleague Sally Rawatton of UNEP and Jesse Slater of UNIDO, who will get us through this afternoon's discussion. I look forward to a rich exchange. Thank you.

Thank you very much, Deputy Director-General. Excellencies, distinguished delegates, dear colleagues, friends, good afternoon and welcome to Cluster 1 of the first Global AI Dialogue, which is focused on the opportunities and implications of AI across its social, economic, cultural, ethical, linguistic, and technical dimensions. It's quite a packed agenda already, but I hope you'll forgive me as the UN Environment Programme's Chief Digital Officer if I point gently to one dimension that is not named in that list, the environmental dimension. AI will shape our economies, our societies, our institutions, but it will also shape and be shaped by our energy systems, our water systems, our mineral supply chains, our waste streams, our climate goals, and our planetary boundaries. There is no truly inclusive AI dialogue if the planet is only present in the footnotes. AI gives extraordinary possibilities to the environment too, from monitoring methane emissions, biodiversity loss, deforestation, pollution, and climate risks at scale, to strengthening early, early warning systems and helping countries design better policies and leapfrog gaps in capacity. But we also need to be honest about the risks. AI has a footprint. Not just energy, although energy matters, and not just data centers, although these matter too. We need to understand the full end-to-end environmental footprint: critical minerals, manufacturing, water, electricity, e-waste, and the rebound effects of using AI everywhere simply because we can. And we need to measure this scientifically, not through slogans or emotive marketing claims. But through shared methods, transparent metrics, and evidence that governments, companies, researchers, and civil society can trust. Only then can we begin to agree on what sustainable AI actually means. And let me add one point that is sometimes forgotten: sustainable AI is also affordable AI. Lean models, efficient infrastructure, renewable energy. Circular hardware, open standards, and frugal applications are not only better for the environment; they are essential if AI is to serve countries and communities that cannot afford waste. And here too, opportunities for innovation abound. At the UN Environment Assembly last December, member states adopted a resolution on the environmental sustainability of AI. UNEP is now taking that mandate forward, and we invite all stakeholders here to. work with us on implementing it. So my hope for today is that as we discuss AI's social, economic, and cultural impacts, let's also remember that the environmental dimension runs through each and every one of them. And my hope for next year is that the environment is named clearly on the agenda, because the future of AI cannot be separated from the future of the planet. Thank you and welcome to the session.

Thank you, thank you very much, Sally. So now we move to the opening remarks by our thematic cluster co-chairs. I'd like to welcome to the stage His Excellency Mr. Marc-Alexandre Doumba, the Minister of Digital Economy and Innovation for Gabon, and Mr. Rasheed Khan, the co-founder of Yellow.ai.

Excellencies and distinguished guests, hello, good afternoon. The length, depth, and variety of the theme say everything that we need to recognize about the complexity of artificial intelligence, so as is the quality of the participants in this room and across the forum. AI is not just business as usual. It is a technology that imposes on us to embrace change, to redesign systems, to create new incentives and to set new measures of success. At a time where individuals have net worth that exceeds nations, it is critical that we engage on a new path. On the one hand, this is a technology that has the ability to amplify our strength and to shore up our weaknesses. We can structure accumulated knowledge, generate new insights from new ways of cross-pollinating data and information that once resided in silos anchored in different languages, context, form, and structure. There is the ability now to convert unstructured data into structured data, which can benefit developing countries disproportionately Since most of our knowledge is tacit and unstructured, we can improve knowledge transmission across generations in ways that were not possible before. If you accept the premise that development is about the accumulation of knowledge, know-how, and capabilities, then AI will certainly stimulate the economy in novel ways. Higher productivity, faster time to market, faster rate of conversion from ideation to market. Entrepreneurship and the process of creative destruction can reach new heights. From a cultural and ethical standpoint, while it is true that AI is largely shaped by Western culture and values, the big opportunities that will come from local and regional adaptation, from big AI to smaller AI, recognizing that the cognitive economy only works if you meet people where they are cognitively and where there is relatedness of thoughts, because that is where trust really builds. From a technical standpoint, much of the opportunities will come from doing AI differently, and that is for 2 reasons. Advanced economies can't sustainably keep this up. Middle and low-income economies can't even aspire to doing AI in the way the big economies are. So there is a need for AI developments that will cater to Africans, to Latin Americans, to other Europeans and people from small islands and people from the Caribbean. Energy will need to be sourced and supplied differently. Water will need to be sourced, supplied, and actually reused. Data centers moving to space will need to be regulated in a way that benefits everyone. And this goes to show how far human prowess can go. As a collective, we are only limited by our imagination. Let us use our imagination to distribute the benefits more fairly, Not more for fewer people, but more for everyone. There has never been a time when there's been as much capital, talent, and technology as we are today. We have absolutely no excuses to do things that will not work for everyone. Thank you so much for your time.

Excellencies, Distinguished delegates, colleagues, a very good afternoon. It's an honor to co-chair this cluster alongside His Excellency, Mr. Minister Dumba, and to open a conversation that sits at the very heart of this dialogue— not whether AI matters, but how its benefits actually reach people. I come to this room as a builder. Over the past decade, I have watched AI move from research labs into hospitals, classrooms, farms, and government service counters, often in dozens of languages across very different national realities. And that experience has taught me one thing above all: the gap today is not the gap of ambition or principles; it is the gap of practical mechanisms. The technologist Tim O'Reilly gave our industry a simple test: create more value than you capture. I can think of no better test for this dialogue. Collectively, the value AI generates must flow broadly than it is captured across our workers, languages, regions, and generations. That is precisely the mandate this cluster carries. In, in line with General Assembly Resolution 79/325, our task over the next 3 hours is to move beyond high-level governance principles towards the concrete conditions, cooperation, standards, infrastructure, the skills that translate AI's productivity, scientific, and sustainability gains into tangible development outcomes. We will examine this through 5 interconnected lenses, first being social, to see how AI serves education, health, agriculture, and public services. Second being economic, What a just transition looks like for workers, entrepreneurs, economies where access to compute, data, and capability remains deeply uneven. Third being the whole cultural and linguistic aspect. How do we ensure AI reflects the world's diversity, including low-resource and indigenous languages, rather than homogenizing the whole aspect of AI? Fourth, the whole technical aspect, the design choices, the standards, the accountability, The mechanisms that determine whether these systems can be trusted, audited, and adapted locally. And finally, as Sally said, the whole environmental aspect of AI— AI's dual role both as a climate tool as well as the growing footprint we must measure and mitigate. And no single actor can answer these questions alone, and this room reflects that. The cluster is the product of broad interagency effort across UN system, co-led by the UNIDO and UNEP, through the Interagency Working Group on AI. And it convenes member states, industry, academia, civil society, and technical community on equal footing. That breadth is not ceremonial. It is the method for us to do things here. So my ask as we begin this, bring us specifics. What has worked in your country, your sector, your community? and under what conditions? Those are the answers that we will carry forward into the Dialogue of Dialogues and onward to 2027. Let us make them worth carrying. Thank you, and I look forward to the discussion.

Thank you very much to Minister Dumba and Mr. Khan. And now to kick things off, it is my pleasure to invite His Excellency Alaa Kariis, the President of Estonia. and Secretary General of the International Telecommunications Union, Mr. Doreen Bogdan-Martin, to the podium for a fireside chat.

Thank you very much, Good afternoon, Excellencies, ladies and gentlemen. Mr. President, it is such a great honor to welcome you back. The President joined us last year at the AI for Good Global Summit. And, Mr. President, since then, AI has evolved at an extraordinary pace. Today, it is becoming a defining force shaping economies, societies, public services, and geopolitics. And as Minister Dumba has just mentioned, it's kind of forcing us to embrace change. As we begin this first thematic discussion of the Global Dialogue, and as we just heard from the co-chairs, I thought we should focus on not just AI's potential, but also its broader implications— the social, economic, cultural, linguistic, ethical, and technical. And at the ITU, we often say that AI will only succeed if it remains centered on people. And I think few countries have demonstrated this as clearly as Estonia. So, Mr. President, let me begin by asking you to reflect a little bit on the broader global picture. And when it comes to how governments and societies are adopting AI, if you could tell us, looking a little bit ahead, where do you see the greatest opportunities, not only for Estonia, but for countries around the world? And a second part of that question, if I may, what do you believe will distinguish societies that are able to harness AI successfully from those that risk being left behind? Mr. President.

Thank you. Thank you very much for inviting me back to Geneva. It's my pleasure to be here and share some of our experiences also from the past and also today. So AI is just not only a technology. It's something that we should rethink how our societies work. And we do have certain experience from past digitizing our societies, and this is something I can probably share also later because there are so many overlapping things and some other issues we could We could compare, but of course I would like always to talk about the positive sides. Of course I do know there are also negative sides and there are a number of risks and a number of things we have to keep in mind and we heard already a minute ago that there are also environmental issues and so forth. So that's why I'm saying we have to change how our societies work. But it does help, of course, entrepreneurs create new value, make governments more responsive, strengthen resilience, and so forth. And but the main thing or main issue using AI is not to be and use it first and fast, but use it wisely. This is something we are we are also talking back home. So We don't want to be first, but we want to be wiser users of this technology. And— but to succeed, of course, there are a number of things we should keep in mind, and I think a part of technology is actually trust. It's not the trust to the technology, it's the trust to the society, trust to the government, trust what we are doing every day. If you don't have trust, I think this won't work or even won't start, to be honest. And of course, how we use these data, how— because going back to our digital society, the data protection was extremely important. And in Estonia, all these data, they do belong to citizens. We don't belong to government. We belong to citizen. But any particular citizen can actually trace who has been interested in the data, either it's GP, either it's police. So there is a trace, and you can follow and ask the question, why did you use my data, and what did you do with that? So I think this trust is important. And transparency of data use and everything else. And of course, the second thing is skills. We need these skills, and not only that the kids understand what's going on, but also the whole society, because if we don't do that, it happens that we generate inequality in society. That means some people have certain advantages, some people don't. But this is something we don't know and it is probably especially important if you look at the bigger countries in developing countries that we do have an access to this technology also in remote areas. I understand it is not that easy because in some of these regions even electricity is not present. So, but we should understand from the very beginning That's important. And of course, the same digital infrastructure. That means we should have this digital infrastructure in place to use all these benefits of AI. But of course, there are risks. And I always give one example from my own past. I used to be a molecular gene technology person. I did biotechnology, and the same situation was actually something like in 1970s, 1980s. People got very scared. Scientists got very scared because they started to imagine what one can do if we modify our genes or genes of plants or whatever. But what happens actually? But where are your rules and regulations is going to be present, and and also. technology has its certain limits. But of course, the difference is that in that time, only a couple of— not a couple of, maybe a couple of dozen people were able to use this technology. But AI is in everybody's pocket, is in everybody's computer, and it develops so fast. And the problem is not technology itself, but the speed. So that's what scares actually people. What's going to happen next? But coming back, we should educate our people that they know what are the risks and how to use this technology, or even importantly, when not to use this technology.

Thank you. Thank you so much, Mr. President. I think wise words from our President who said we need to use artificial intelligence wisely. I think well said. I want to pick up a little bit more on the skills and education piece. Our ability to prepare citizens with that right knowledge and skills is going to determine whether artificial intelligence becomes a force for inclusion. We spoke a lot about that this morning in the plenary, as well as a force for innovation, or if it becomes a source of new divides, especially when it comes to education and the next generation. So, Mr. President, that's what I wanted to get you perhaps to comment a little more on. We spoke previously about Estonia's AI leap, and many of us know Estonia has had many leaps. And so, if you could share a little bit in terms of that AI leap that really attracted international attention in terms of helping schools to best prepare for the age of artificial intelligence, where you have managed to support both students and teachers. And a lot of countries are asking, you know, how did you do that? How did you make sure that AI didn't become just simply another classroom tool? And how can we make sure— and this is a question we hear often— that with artificial intelligence, we can still nurture creativity, we can nurture critical thinking, and keep that curiosity flowing? And then the last one— this is a 3-point question here— what lessons could you share, Mr. President, that could be valuable to the many countries we have here? I think we have 170 countries that have picked up their badges. So what lessons could you share with the countries here that might be at different stages of digital development?

Obviously, I am biased because I used to be a professor. That means education is number one as far as I can see when development of different countries is concerned. And, well, actually, a couple of years ago, I gave a speech back home. We had an anniversary of our country. And I mentioned AI, and that means that we should do something about AI. People didn't pay much of attention, to be honest. So I decided to invite entrepreneurs, entrepreneurs, also educational people and government people to discuss what, what should we do. There are a number of options because AI is everywhere. So where should we start? Where should we focus? And we decided that education is the key. And we started to educate, first of all, teachers. I mean, in our country, there is— anyway, there is a lack of teachers. But we need also skilled teachers on AI. So we made courses. Basically, every teacher could get courses last summer to learn what the AI possibilities and opportunities are, risks as well. And we started with upper secondary school. That means 9th, 10th grade, and 11th grade. The reason was very simple, because in 2 years' time, they are going to a university. That means they should have these kind of skills, how smartly or wisely use this AI. And of course, the same with the teachers. And now it's 100% of teachers and same of students are familiar what are the possibilities. And I always say it's not compulsory, but you still can use at school chalk and blackboard, but teachers should know what are the other options. That means to diversify how they actually teach at school because the kids They pick up this technology very easily. And what we did, the second thing was, we are not able— we're a small country, we're not able to build our own platform. So that means we collaborated with OpenAI and Google. So they made a special platform for Estonian schools. It's not that they give you a simple answer if you're something, but they start discussing with you. That means make you more think what you want to get. Because when I was young, there was only one right answer, the answer what was given by the teacher. At least the path to this result should be the same. But now there are different paths. But when you get these different paths, you should also be able to discuss and why you choose this one or that one. That means to start— you mentioned critical thinking. It's very important because we see nowadays it's a problem. It's not a problem with kids. It's a problem with politicians, with everybody. They don't have this kind of critical thinking. And I do believe that AI actually does help to develop this critical thinking. So there are so many things. And we started, as I said, upper secondary school, but now we also go to primary schools. And it's also— for primary schools, it's not compulsory. And when we started, a number of teachers were not very happy. A number of students say they are never going to use it. But I have another example. I got a friend when digital cameras emerged. He said, no way I'm going to use digital camera. No way. A couple of years later, he had 2 of them. So that means the technology develops and the same happens. Different people have different pace. So, and you have to accept it. So, and as I said, it's not compulsory that you use technology, but you have to know how it works.

So I'm just gonna kind of follow on that sort of same line of thought. You have, made AI literacy a national priority. I think you set a target of having 100,000 people acquire practical AI skills. And I think that's linked to the platform you mentioned, the sc.ai initiative. Tell us a little bit what that impact has been of making AI literacy a national priority. And also coming back to advice for this crowd. What advice would you offer governments that are seeking to ensure that their workers, their entrepreneurs, their public servants, and all their citizens have the opportunity to participate in and benefit from the AI economy?

I, I maybe— well, maybe I'm not the right person to give advice to governments, but nevertheless, I think the best advice would be that should develop curiosity in students and also that one should not— and this also applies to adult people— one should not be afraid of unknown. So that means don't be afraid. There are so many options, different options in the world. Go have a look, be curious. So this is what I would like to say to to actually to all, to all people. So curiosity is a driving force to my mind. But what was the other questions? I already forgot. I need probably AI to remember things.

So, well, I think you, you've touched on the main points, but it was like in terms of the job market, if you wanted to give countries some tips about how AI skilling can impact Entrepreneurs, their job markets?

They are not going to take away your jobs. They just leave more time instead of doing stupid things, doing some smart things, because you don't have to have this kind of bureaucracy, this, you know, we even have in Estonia a company now which uses AI to make contracts between between different companies. Only AI. Of course, these companies, they don't know, but they are satisfied. So that means AI is— because they are very trivial contracts. I mean, there is no— you don't need lawyer. But AI does this kind of job very well. So just you have to change how you think, you change how you do your work, and probably have also much more time for for holidays as well, and time for thinking.

That would be nice, wouldn't it? More holidays. I think some of us might need it after this week. So, ladies and gentlemen, you have heard it. As, as the president said, we must be curious. Let's be wise in our use of artificial intelligence. And I really want to Thank you for being back with us this year and also for reminding us that the future of AI will not be determined by technology alone. It's going to be shaped by the choices that each of us make, how we invest in people, how we build trusted digital foundations, how we foster innovation, and how we cooperate across borders, across sectors, and how we also learn from each other. And, Mr. President, I want to thank you so much for sharing your thoughts and for sharing the experience of Estonia. If you'd like to share a last word.

I do, because I forgot one important thing, which is language. I'm coming from a small country. That means language is extremely important. That means we have to develop all these platforms that they understand Estonian language in our case. And that means they have to have an access to literature, I mean modern literature, newspapers and so forth. It's not easy, I know. There are IP issues and so forth. But recently one of the main newspapers gave an access to our one of our institutions to use all the articles that they have been publishing already from 19th century till up to now because we need modern language. It's not enough that you read old books and use this language. So that's important for a small nation and Estonia is not the only one. I mean, there are plenty of small nations in the world and Because otherwise, otherwise, we lose the language because they switch to English, which is very convenient for young people these days. So keep the language and culture alive as well using AI. So these are my last words. But thank you very much for inviting me here.

Thank you. Thank you again, Mr. President. More wise words— keep the culture and the language alive. Thank you so much for this conversation. I think this helps to provide some additional food for thought for this thematic cluster. With that, thank you, and back to our co-chairs.

Thank you very much. A round of applause. And now I'm pleased to announce that we will be transitioning into the panel segments. So we would now like to invite Ms. Mary Robinson, the former President of Ireland and former UN High Commissioner of Human Rights, to introduce our first panel of the day on AI for inclusive development and social impact. Please come to the stage. Round of applause.

So I have the pleasure of introducing a wonderful panel on this thematic cluster 1, AI opportunities and implications, social, economic, cultural, ethical, linguistic, and technical dimensions. We have 25 minutes, so please, panelists, come up pronto, um, as I introduce you. Um, and I'll begin with Yossi Matthias, Vice President and Head of Google Search. Jamila Venturini, Co-Executive Director of Derechos Digitales. Lan Shue, Professor and Dean at Tsinghua University. Lasina Kone, Director General and CEO of Smart Africa. And Leslie Cho, Senior Director of AI Products at AI Singapore. Okay, and I'm going to ask you questions in the order in which I introduced you, um, because that makes it easier for me. And I'm going to begin with you, Matthias, um, Head of Research at Google. Um, how can we build on breakthroughs in large-scale AI to ensure systems reflect linguistic diversity, local knowledge, and regional realities across all communities? I mean, the President of Estonia spoke about the importance of preserving our language and culture. So how do we make sure we have this inclusiveness and diversity, local knowledge, regional realities?

Thanks for the question. Mic. Thanks for the question, and thanks for having me here. Indeed, the last comments made by the President of Estonia really resonated. As many of you may know, the mission of Google is really to organize the world's information, making it universally accessible and useful. So this notion of how to make it available in all languages has always been a prime consideration. Now with AI, we see stark reality. About nearly half of the training data of major AI models is based on English. English is only 20% of the spoken spoken languages. Now, we make progress on these areas in, in, I would say, 3 pillars. One is overcoming data scarcity through ML, machine learning breakthroughs. The other one is about anchoring on data collection directly on local communities. And the third is about designing for deep cultural nuances. On the first one, we do have a bold objective to have a 1,000-language initiative to have a zero-shot, zero-resource machine learning, the way we call it, on how to actually train machine learning models, and leveraging on breakthroughs in machine translations with AI to actually build through languages. And recently added 110 new languages to Google Translate, including 60 African languages. By the way, Google Translate just marked 20 years now with supporting of 250 languages with 1 billion users per month. A brief note on the second pillar on anchoring data collection directly on local communities. We have ongoing efforts in Global South, including in Africa, on how to source spoken language so that we can actually make them available as open source in order to help training models based on them. And related, we have also in India efforts with 150,000 hours of real-life speech across 773 districts of India, open-sourcing dataset local developers and researchers to build on. So AI is a lot about making sure we have the right data collection represented for those who are building the models. And a brief note about the third one, which is local cultural nuance. Obviously, this is highly important. So first and foremost, it's about developing benchmarks. You know, a lot of— with AI revolution, a lot is about how we validate and how we actually solve for the benchmarks that we want to optimize for. And we are designing these benchmarks. We're actually developing our platforms and are encouraging others to actually use those benchmarks. And I would just note that with Gemini, we expanded this year to more than 70 languages across more than 230 countries, making Gemini, which is Google's AI model, the most widely available AI assistant in the world. Last but not least, this notion AI is evolving, and I think it's extremely important that we are making progress on understanding the cultural nuances and making sure that they are reflected in both the training data and how we evaluate our models, and making sure that the experience is really solving for the different communities that we have around the world. Thank you.

Thank you. Those are impressive numbers. So, I hope we are closing this digital divide and this AI divide in our world. The next question goes to Jamila. And you're co-director of Derechos Digitales. And I'd like to recall what the UN Secretary-General said a month ago, but also repeated this morning in his speech. It was a challenge to AI companies on their environmental footprint. How can that footprint— energy use, water consumption, e-waste— be measured, disclosed, and mitigate it while harnessing AI to accelerate climate action, biodiversity protection, the circular economy, and other environmental priorities? Thank you

so much for the question. This is a pleasure to join in this conversation. To start, we cannot talk about environmental— mitigating environmental impact without challenging the idea that AI is inevitable. That is just not true. AI development and deployment today is being driven mainly by the economic interests of a few companies and countries, which have been shaping both technologies and governance processes until the date to protect those same economic interests. AI is not an autonomous force driving itself, at least not yet. What we call AI implies a supply chain that goes from mining to data processing that requires an immense amount of natural resources to operate. And we heard a bit of that in the introductory remarks of this panel. AI's environmental footprints are structural. They are not accidents. And they often impact workers, communities, and territories that are far from the main beneficiaries of AI deployment. Global South countries in particular are at the same time providing those resources and facing the consequences of such impact. Impacts without obtaining any significant gains from the AI economy. Addressing the environmental impacts of AI implies serious commitments from multiple stakeholders, and I need to discuss the conditions under which the AI is developed and deployed. On the one hand, we need to integrate the principle of precaution into the global framework for AI governance that we are discussing in this event. On the other, such a framework needs to advance concrete commitments to transparency, accountability, and redress at the transnational level, making sure to meaningfully engage civil society, particularly from affected communities, if we want those mechanisms to be effective. We need binding commitments with transparency regarding AI and mandatory obligations for disclosure of resources of consumption. For instance, water and energy, and supply chain level impacts. Those can build from existing frameworks such as the Escazú Agreement, which is the first environmental treaty of Latin America and the Caribbean that promotes the rights of access to information, participation, and justice in environmental matters. Addressing environmental impacts, labor, and broader human rights impacts within the transnational AI supply chains is an urgent task for this dialogue. While the work of the scientific panel is crucial to provide evidence to policymaking, documentation of impact cannot be restricted to academic consensus, as scientific consensus demands time that communities that are today being impacted by AI don't have. We need to support sustainable mechanisms for independent documentation that leverage civil society capacities and ensure that existing UN human rights mechanisms, including universal periodic reviews, special procedures, treaty bodies, and the Human Rights Council, are adequately resourced and equipped to monitor, document, and respond to AI-related harms. And finally, we need to strengthen coordination within other international processes and institutions. We can mention IPCC and also the ILO to ensure the subject is not addressed through a siloed AI-specific mechanism. We need to be clear that there is no easy fix when it comes to these complex issues. Beyond promises of efficiency, there is little evidence that AI itself can promote environmental protection. What we already know is that more advanced models have increased footprint. The impacts of AI at the supply chain level need to be part of this conversation, and it is important to have in mind that more AI or better AI cannot resolve impacts created by AI themselves. Thank you,

Jamila. And it does seem to me that this is something that civil society in different countries should really rally round and insist on this transparency, insist on implementing that call to action from the Secretary-General. My next question is to Professor Lan Shuai. Um, I'd like to ask about what forms of international cooperation can support open data and open AI models across the public sector to prevent dependency on a handful of providers. Thank you,

Madam Chair. I, I think that this is a wonderful question. I think that indeed, I think many, uh, participants in this conference would be very interested in in understanding, you know, how to do that. Let me make 3 suggestions. I think the first is to protect the open global AI ecosystem. I think the open global AI ecosystem allows various AI frontier models, open-sourced or closed-source, to compete fairly in the market and to allow global users to enjoy the benefits. technological advancements and the competition. For example, we see the— as our colleague from Google mentioned about the frontier models, Gemini and many other frontier models. I think there are many— for example, there are several expected IPOs of the US frontier companies such as Anthropic and OpenAI with valuations around hundreds of billions. At the same time, we also see the global reach of Chinese AI open-source models have undergone also a quantum leap that becomes an indispensable force in the global AI ecosystem, reflecting not only the download statistics but also in the widespread adoption by developers, enterprises, and national strategy worldwide. For example, in the past year, over 40% of the global open-source model downloads came from China, from Chinese open-source models. And also, Chinese open-source models powered nearly 30% of global AI usage. And also, there was an estimation that about 80% of the US AI startups use Chinese OpenAI source models in their pitch to various venture capitalists. The core drivers are really high performance and low cost. These open-source models support digital sovereignty, allowing users to keep data locally. They also embrace local languages. Models like Q1, they support over 100 languages providing a foundation for AI developers in Africa and Southeast Asia that have long been overlooked. I think now there is also some discussions about how actually those open-source model and frontier models can work together. I think that's indeed— I think the discussion indeed, particularly with regard to safety issue, That needs to be discussed and needs to be worked out. But certainly, I think to protect the open global AI ecosystem is ultimately— it's extremely important for maintaining the reach and the adoption globally. The second, I think, is to cooperate on capacity building in AI. I think that capacity building is really a potential area for international collaboration. I think the development and deployment of AI requires both physical infrastructure and technical capacity that many developing countries are lacking. So in order to address this challenge, I think UN had adopted a resolution in 2024. I think that should be really worked on. And the third is to work together to build guardrails to prevent AI risks. I think that many colleagues here are very familiar with that. And there was also, most recently, there's a report, I think, by the AI Sentinel panel. So I won't elaborate too much on that. Thank you. Thank you

very much. And I'm glad you mentioned the 2 things, capacity building, and I understand there's been a commitment on that at this meeting, but also the guardrails that are necessary. I now turn to immediately on my right, the Director General and CEO of Smart Africa, and I understand that your initiative is training tens of thousands of policymakers in AI and digital skills across the continent of Africa. From your perspective, what determines whether AI deployment becomes a genuine engine of local job creation in developing regions versus another wave of disruption African economies simply have to absorb? Thank you

very much for inviting me. I believe we often focus too much on AI itself and not enough on what's around it. From where I sit, working with the 42 African countries, the real question is no longer who has access to AI, The real question is who has the capacity to shape it. Today, powerful AI tools are increasingly available everywhere, but access alone does not create prosperity, Madam. Access alone does not also create jobs. Access alone does not create values. The countries that will benefit most from the AI will not necessarily be those who use the most AI. They will be those that build the capacity to adapt, govern, and create with AI. For Africa, this means, you know, asking simple questions. Are we participating in AI economy revolutions, or are we simply consuming the product of AI economy. Because if we only import AI solutions developed elsewhere, then most of the intellectual property and most of the innovations and most of the jobs will be created elsewhere. The greatest risk for Africa, Madam, is not that AI moves too fast, but the greatest risk is really that we remain spectators while others capture the value. And this is why, precisely why, Smart Africa has made skills development such a priority. I often say that AI will not replace people. I'm speaking close to you here. I've got to be careful. And that people using AI will replace people who are not using AI. But skills alone, Madam, is not enough. We are engaged in that, of course. The AI succeeded when it's supported by enabling environment. It requires connectivity. It requires trusted data system. It requires digital identity. It requires effective institutions. It requires governance. It requires trust. It requires trust. That's why we advocate always a system approach to digital transformation, Madam. AI cannot be discussed in isolation. It must be built on the foundations of digital public infrastructure, DPI, trusted data ecosystem, as well as human capacity and good governance. At the same time, Africa's opportunity here is not necessarily to compete in building every frontier model. Our opportunity is to apply AI where it matters most to our citizens. And I say that all the time, we don't need AI girlfriends or AI boyfriends. We got to go to the necessary stuff, essential agriculture, improve productivity, food security, healthcare, expand access and improve outcomes in education and to personalize learning. in government services as well to improve efficiency, transparency, and inclusion. I think we must remain citizen-centered. People do not wake up, Madam, in the morning asking for AI. They want better healthcare, better education, better jobs, better public services. AI is simply a tool, a tool that achieves those outcomes. So the future belongs not only to those who build the AI, but also to those who can adapt AI to local reality. Languages— we talked about it in the morning in the panel. We have over 2,000 languages in Africa. Language matters. Culture matters. The President of Estonia said it. Context matters. We are from Africa. We are already seeing examples where locally adapted solutions outperform imported ones because they better understand local needs. And we've been there, you know, leapfrogging with the mobile money. People didn't believe in Africa. Today, Africa controls $1.4 trillion in transaction value in mobile money. So how many models will be used? AI in Africa will not be measured by how many models will be used. No, it will be measured by how many jobs we create, how many businesses we grow, and how many citizens will live into the digital economy. Thank you very much, Madam. Thank you

very much for that powerful statement. Your point is very well taken, that access alone is not enough. You need the capacity to adapt. To local realities. Absolutely. I now turn to Leslie, Director of AI Products at AI Singapore. And because you're someone who builds and deploys AI, I want to ask you, if you were advising a minister in a middle power or Global South country who has a limited budget and a narrow window to act, what action needs to be taken now? Good afternoon.

Good afternoon. Before I— I have 2 responses to the Minister's question, but before I do so, I just want to emphasize that some of what I will say was not synchronized but builds very strongly on what was said by His Excellency the President of Estonia and my colleague Bless you. Here on the left. First of all, access is not the main problem. It's necessary but not sufficient. But my answer to the minister will be, one, don't spend your budget trying to win the compute or infra game. You're not going to win it, nor is it necessary. Spend your energy and budget on what my colleague here just said, the enabling environment, the complements. And let's not— you know, we like to focus on the expertise, the technology. I like to build models. I like to be very technical. But really, this is a human problem. It's a societal problem. It's about relearning skills. Reinventing workflows, changing our society. To me, it's ultimately a leadership problem. We and you will have the challenge of thinking, how do we make the environment— how do I inspire my people to build and use this technology in a trustable way doing useful things while managing all the risks that come with this technology. The second thing I would say to the Minister is don't underestimate what's already there for free. It is amazing what you can get on YouTube, and thanks to tools Like Ollama and Open Models, a student with a laptop can have access to powerful AI, what we would call frontier 2 years ago. So yes, concentration matters, cost matters. The divide is real. But I want to point out that the glass is also half empty. Yes, we point to the empty side. I want to point to the water, right? What can you do with what's there? And it's amazing. We didn't synchronize this. Mobile payments. Yes. Mobile itself. It was not a frontier technology when it hit the emerging world and the Global South. And yet, the Global South made these 2 technologies really, really useful. So again, not to diminish many of the challenges, but the opportunities are real. And if I may, the last question I would sneak into the minister is the following. The gains and the potential of AI is great, but Don't underestimate distribution and transition. We made this mistake with globalization. The theory was right. The gains were real. But because we miscalibrated on distribution and transition, and we thought the invisible hand will take care of everything, We now face many more challenges today. If we want AI, if you want AI to be useful, to be trusted, to be managed appropriately, we have to think hard about distribution and transition costs. Thank you. Thank you very much

for that, and thank you for the link back to the mobile money transfer as being Very crucial. I only have time for one more question because I've been warned we're very tight to time, but it is, I think, an important question, and I'd like to put it to Jamila. What can be done concretely to prevent manipulative and unsafe design patterns across AI systems that shape people's decisions, behavior, and access to information? What transparency and independent research— researcher access are needed to make oversight effective? And you only have a very short time to answer that question. I will do my best. Thanks

for such an important question. Manipulative design facilitates the exploitation of the people who are already most vulnerable in face of these systems, including children, elder people, and other groups that have been historically marginalized and excluded from technology access and literacy. At the same time, impacts goes much beyond individual level. It generates collective and structural harms, including affecting our democracies, exacerbating inequalities and discrimination. 3 things I believe are important to highlight in this regard. The first is that manipulative design is not a bug or restricted to complex systems. It combines social engineering with technology and is embedded within business models that depend on data and attention to generate revenue. 2, as there are economic incentives for certain types of manipulation, self-regulation has proven not to be sufficient to prevent abuse. And 3, yes, digital literacy, data literacy, AI literacy are important, but they cannot become an excuse for impunity. And placing responsibility only on end users is not enough. Responses require addressing the structural conditions that incentivize such manipulative design, and they range from data protection to platform regulation. Let me give a quick example to illustrate what I'm saying. It is a fact that synthetic images generated by AI, what we usually call deepfakes, disproportionately affect— disproportionately generate non-consensual sexual content, and that most people represented in them are women, 99% according to data in 2023. So, these platforms allow the expansion and facilitate the expansion of gender-based violence. A study by a Latin American researcher working with Derechos Digitales has shown that among more than 100 apps freely available at the Google Play Store, 70% depicted sexually suggestive or nude images of women as part of their promotional materials. This is a design decision. Is not casual. And in this case, the violence starts from the fact that those platforms are easily available for download. Some concrete recommendations: manipulative design presents a serious threat to equity, agency, autonomy, and dignity, which are foundations to human rights. Regulation and oversight are necessary to prevent abuse, and mandatory obligations in terms of transparency and access to disaggregated data should be established, including in terms of reporting when risks are detected. Finally, and most importantly, systems should be, should be subject to human rights impact assessment throughout their life cycles and should face restrictions to their circulation in case providers cannot prove that they are safe for all types of users. Let me recall that General Assembly Resolution from 2024 already recommends states from refraining from and ceasing the use of AI that poses undue risks to human rights, especially from vulnerable groups. And I believe we should build from these recommendations as we move forward with a global standard regarding manipulative design. Thank you again for the question. Thank you very much, Jamila.

I think this panel has done its best in a very short time to cover both the opportunities and also the implications, some of them negative implications, of the social, economic, ethical, cultural, linguistic, and technical dimensions. And we do again thank the President of Estonia for being quite inspirational to this panel. And please give this panel a good round of applause. Thank you. Bye-bye. Do we have your names up? Good

afternoon, Excellencies, ladies and gentlemen. We will now facilitate interventions for this segment from the floor. Statements will be delivered on this based on the speakers list that was established through inscription that was made available on the UN Global Dialogue on AI website. Please note that in the interest of promoting broad participation from all stakeholders, we will proceed in the following manner, with one representative from member states followed by one representative from stakeholders alternatively. The speakers will be invited to intervene from the lectern on the stage. Speakers, can you please come to the first row so that it is easier for you to come to the lectern to speak? The e-delegate list is closed and any changes has to be communicated to the secretariat in the room next to the podium over on my left, your right. A timer has been set and is visible on the screens as well as on the lectern. We kindly encourage all speakers to respect the allotted time so that the largest possible number of speakers may be accommodated during this segment. Microphones will be automatically switched off once the allocated time has expired. We therefore respectfully request speakers to conclude their remarks within the time limit. Thank you so much. And over to the co-chairs. Thank

you. Distinguished guests, from Finland, allow me to invite to the stage the Undersecretary of State for International Trade, His Excellency Jarno Sijarla. Thank

you. Thank you for the speakers and panellists for setting the scene for this discussion. Finland would like to take up on 3 points we believe that should be be further discussed in the framework of the Global AI Governance Dialogue. All these points were addressed also in the preliminary report of the Independent International Scientific Panel on AI. Finland values highly the sharp analysts delivered in the reports, and this— and it resonates with our understanding of the priorities ahead of us. First, Finland underlines the importance of open connection, open competition and innovation as the independent scientific panel report rightly highlighted. There is a high degree of market concentration of AI chips, computed models, computer models, which are largely in the hands of the few entities globally. The economies, customers of the present market dynamic are not optimal, and this concentration also poses risks of economic security by creating strategic dependencies and vulnerabilities in the critical supply chains. We need to remind ourselves that the Global Digital Compact goals for tackling concentrations of technological capacity and market power, and pursuing more equitable, inclusive in the digital economy. The key discussion point for the AI dialogue is therefore how global AI governance can contribute to creating a more level playing field and business environments where businesses of all sizes and from different parts of the world have fair opportunities to respectfully Compete, and innovate in different parts of the AI stack. Secondly, Finland emphasizes the importance of global AI governance. Okay. Thank

you very much, Your Excellency. Allow me to invite the CEO of Hey IT Solutions Limited to the stage, please, for your remarks. No? Okay. I would like to invite the Deputy Minister of Foreign Affairs of Russian Federation onto the stage, please. Dear colleagues,

Russia welcomes the launch of the Global Dialogue on Artificial Intelligence Governance as a dedicated United Nations platform for discussing cooperation in this field. We note with satisfaction the efforts made by UN member states and the UN Secretariat as well, by independent international scientific panels to agree on the modalities and objectives of multilateral engagement. We call on delegations to associate themselves with the statement of the Group of Friends on Artificial Intelligence Capacity Building delivered in the Plenary this morning by our colleagues from Zambia, and ask the co-chairs of the Dialogue to include its text in their summary. It has taken us quite some time to arrive at today's meeting and to prepare for it, moving from behind-the-scenes and semi-closed consultations among a narrow group of states and their monopolies towards fully-fledged multilateral discussions involving diplomats, scientists, and developers. A significant contribution has been made by India, which undertook to organize the Artificial Intelligence Summit in New Delhi this February. We call on— we call for the outcomes and conclusions of that summit to be built upon in further work under the auspices of the United Nations. The Russian Federation is committed to addressing the challenges of development and security in the field of artificial intelligence, bridging the digital divide, raising labor productivity, ensuring non-discriminatory access to technologies, computing capacities, and the data that powers them, and developing open-source computer models while upholding security requirements and human rights. It is essential to ensure that developers of artificial intelligence systems comply with the laws of the countries in whose territory such technologies are applied. We have consistently advocated the creation and deployment of trusted artificial intelligence systems in order to ensure an adequate level of protection. We regard as the constant goal of our joint efforts the building of a fair and equitable system of international regulations of the use of artificial intelligence technologies, one that is consistent with the principles of the UN Charter, above all those concerning respect for the sovereign equality of states and non-interference in their internal affairs. The non-state stakeholders taking part in the global dialogue, whether developers or experts, are called upon to assist efforts to build and strengthen trust among countries and peoples. Only such work will serve as a safeguard against digital neocolonialism. Russia stands ready to share its potential, its human resources, scientific and technological expertise, and energy capacities for the development of sovereign artificial intelligence technologies. We have time-tested achievements and solutions at our disposal. As early as the 1950s, research teams in the Soviet Union explored and demonstrated the capabilities of machine learning. Our delegation, that includes representatives of the government, research institutions, and technology developers, intends to share these experiences. We are ready to discuss such cooperation. Thank you very much. Thank you, Excellency.

I would like to call upon the founder and CEO of minaiwater.org. Just a final call for the founder and CEO of minaiwater.org. Okay, we can move on to the next. Allow me to please invite Dr. Emadeline Fatemizadeh, Head of AI Technologies Development and Applications, Secretariat of the Islamic Republic of Iran, to the stage. In the name of God, compassionate

and merciful. Ladies and gentlemen, gentlemen, AI is a test of our values. It will either empower humanity or push people to the margins. In this context, I must express deep concern over the incident of April 6th involving Sharif University of Technology. A significant part of our computing infrastructure, entirely civilian, dedicated only to science and academic work, was damaged. Research in public welfare healthcare, and academic field was disrupted. A university is not a battlefield. A scientific computing center is not a military base. Research infrastructure is not a military target. This shows what happens without effective governance and deterrence frameworks. When science is attacked in one country, scientific security is weakened everywhere. This is why bridging the AI divide matters. AI is legitimate only when it serves humanity, dignity, and justice. We believe AI needs its own broken chair, a symbol standing not against landmines, but against a silent violence of unregulated technology, against attack on science, against erosion of dignity, against a future built without justice. For 4 years, we have moved AI from words to action. New governance, specialized institutions, GPU-based data centers, a national AI platform. We trained more than 3 million students and over 100,000 teachers under Digital Iran project. We support university and startup to solve real challenge— water, energy, and healthcare. Global AI governance must include developing country It cannot be shaped only by those with the largest data centers. Therefore, we propose 5 regional actions. First, a regional AI academy to train talent and connect universities. Second, a regional innovation and AI development fund to finance startups. Third, shared regional computing infrastructure dedicated to science, healthcare, and public good. Fourth, a regional digital free zone to support AI-driven innovation. Fifth, binding legal framework to prevent harmful— Yes, sorry. Sorry. Yes, it's okay. Thank

you so much for your remarks. I would like to please invite the Managing Director of ConsenSys Optimus Incorporated. Good afternoon. My name is Nick Ashton-Hart.

Amongst other things, I am a member of the UN's Data Governance Working Group, which informs my comments, though I'm not speaking on behalf of the group. Our chair, Guilherme, is going to do that later. Thanks to the chairs and the secretariat for organizing such a great meeting. AI is a general-purpose technology. I think we all know this, like the internet, computing, and electricity, and it will transform all of socioeconomic life. The current debate around managing that shift has a strong sovereignty narrative. There is absolutely a role for sovereignty-based approaches, but as they say, you can have too much of a good thing. 20 years ago, the international community had to choose between a single interoperable open internet and a patchwork of national splinternets. Proponents of the latter approach anchored their solution in sovereignty-based approaches, with arguments very similar to those we hear about AI now. Fortunately, the open model prevailed, which delivered open, international, standards-based product and service development and global approaches to governance. That's why anyone with a good idea for a product or service can go global from day one. AI requires global cooperation to an even greater extent than the internet. Let me give just 2 examples. One, socioeconomic assumptions come with biases, which when built into model design flow through to every application built on them. Addressing this requires collaboration from people across many contexts, which sovereignty-based approaches can frustrate. Linguistic and other diversities matter, as we've heard so many times today. Second, overly restricted localization regimes often propose the data sovereignty outcomes can fragment the data that model training depends on, reducing dataset diversity for pre-training and training of models. Excellencies, ladies and gentlemen, no single country can replicate the broad-based technical, practical, and governance cooperation the best AI outcomes require. Even countries where AI development is most intensive, there are the same challenges, including those I've mentioned. International cooperation is necessary, and it has to be multilevel, multistakeholder, continuous, and global. That is how trust will be fostered and solutions delivered. We can and we will compete at every level to deliver innovation and national competitiveness, just as we have with ICTs generally. That doesn't mean we can't simultaneously cooperate to solve shared challenges and deliver the best outcomes from the very same products and services. We can, we should, and the time is now. Thank you. Before I call upon the next participant on

stage, I would request the stakeholders from IEEE and, and GI Ruskind University to be ready on the left-hand side. With that, I would like to call upon The co-founder of AI Safety Asia, onto the stage, please. Excellencies, colleagues, thank you. I want

to leave one message today. The countries that benefit from AI will not simply be those that deploy AI the fastest. They will be those that can practice, adapt, and course-correct fastest. That is why scenario planning matters. Too often, scenario planning is treated as crisis preparation. It is more than that. It is how countries prepare for the opportunities. It is how governments act before the systems are already locked in. What would it take for AI to improve public services? Strengthen small businesses, support workers, protect local languages, and make our economy more resilient. It is also how we ask the harder questions early. What if AI changes jobs before unemployment numbers move? What if cyber attackers use AI faster than defenders can respond? What if a public service AI system works well in English but fails in local languages? What if productivity gains are captured, but workers are left with higher targets and worker support? These are not separate issues. They are all readiness questions. At AI Safety Asia, this is what we see in practice. In our capacity-building work with public officials in Southeast Asia, including senior Indonesian civil servants, officials are not asking only for principle. They need to practice decisions. How to procure systems? How to evaluate them? How to coordinate across agencies, how to manage cyber risks, how to communicate with the public, and how to respond when systems behave unexpectedly. The same logic is behind our Southeast Asian Observatory. Across Southeast Asia, AI policy information is spread across many ministries, languages, and legal systems. Before institutions can act, they need to see the record, verify the sources, and compare options. Evidence is the starting point, but evidence becomes powerful when institutions use it to make better decisions. AI transition may not first appear as mass unemployment. It may appear quietly, So that is something that we need to understand. Thank you. I would like to call upon the stage the Undersecretary

for Communications and Information Technology at the Ministry of Transport, Communications and Information Technology of Oman. Bismillahirrahmanirrahim. First of all, I would like— In

the name of God. Excellency, you couldn't be here at the moment, so I'm just filling in for him. And we would like to, just as a representative of Sultanate of Oman, we would like to share some thoughts with you. I mean, artificial intelligence has been creating lots of opportunities. We have seen it. I'm sure all of you have seen it. But with those come some challenges. Some challenges might be obvious, and some challenges are creeping on us unnoticed. And with all these opportunities, these implications, and they come hand in hand, and we need to deal with both. The cultural linguistic dimension of AI is very important, and as I'm sure you've heard, lots of the panelists today have discussed this point, and it's a very, very important point, we think. You, in your country, you are the best expert in your own culture and your own language. You cannot expect a foreign company to know your culture and to know your language and to know your values better than you. The tools that you need are already out there. There are people in your country, most likely young, but there are people already in your country, they know how to do this. They can create these solutions for your country, for your cultures, and for your languages. We've done it. We have all the confidence that every country on this planet has the capability of doing it. And last but not least, our message to this cluster is a very simple one. Let's embrace the opportunities that AI is giving us with confidence. Let's shape it with wisdom and let this technology reflect the diversity that we have on this planet. Thank you very much. I thank you so much. I would like to invite the Senior Director,

Technology Policy at the Institute of Electrical School and Electronics Engineers, Mr. Audja Caspersen. Mrs., thank you. Thank you so much, co-chairs, for the opportunity to share a few

brief observations during this vital dialogue. As the panel reminded us multiple times this morning, nothing is inevitable. IEEE, like the ITU, traces its origins to the first age of electrification more than 140 years ago. Our experience reminds us that the societal impact of technology is never predetermined. Outcomes follow choices. In that spirit, a few observations on this cluster's guiding questions informed by the scientific panel's preliminary report. First, on human oversight, accountable design, and interoperability, the panel observes that the evaluation of agentic AI systems faces standardization and reproducibility challenges, and that developer-defined risk thresholds currently operate without standardized evaluation or external verification. Governance is strengthened by remaining lifecycle-oriented, from conceptualization through retirement, assessing actual impacts alongside stated intent. Shared standards begins with shared terminology. Without common vocabularies and taxonomies, jurisdictions measuring different things under the same label cannot compare results. Comparability precedes verification. Second, on cultural and linguistic diversity and on participation, the report documents that AI systems today serve a small fraction of the world's languages, while many languages already possess the foundations for meaningful inclusion. Terminology, benchmarks, and test methods shape who is served. Equitable participation in standards development is therefore itself a governance consideration, one IEEE supports through open, global, and inclusive processes. Third, on environmental sustainability, the panel notes the standardized measurement and reporting across the full AI lifecycle remain lacking. Agreed metrics and disclosure practices for energy, water, materials, and e-waste are well within the competence of existing technical processes and a concrete near-term area of cooperation for this dialogue. Fourth and last, on children. It was said earlier in these sessions that AI is reaching well ahead of safeguards. That's a direct quote. Nowhere is this more true than for children. The panel finds that current AI systems can amplify risks to children while deployment continues to outpace development of robust evidence and regulatory safeguards, even as under appropriate safeguards, AI could strengthen children's rights to information, education, and expression. The answer to the first finding is not to deny children the second. It is protection by design. Age-appropriate design can be translated into testable technical requirements demonstrating that safeguards can be engineered in from the outside without constraining innovation, something my organization has been working on for years and happy to share the results of. The same principle should extend across AI. Safety and security are properties of the enabling environment designed in from the start. IEEE stands ready to support the dialogue. Thank you so much. Thank you so much. Thank you. Thank you. And this will be the last

speaker for this session, I would like to invite His Excellency Jevorg Montashian, First Deputy Minister of High-Tech Industry from Armenia. I think he's excited to be the last speaker for this panel. Thank you very much, panel, and I will try not to

annoy you at least, but I want to start with thank words to distinguished chairs, Excellencies, and ladies and gentlemen. Thank you very much, and I want to send my sympathy for today's speech of the UN General Secretary Antonio Guterres about an important subject which was, I think, not that much often we can hear, and this is an important milestone. Thank you very much. for the achievement of this gathering these days. I would like to continue right now what was prepared as a speech. And AI is already— I think no need to prove that it's a tool. We are listening very often that seems it's a little bit questioned by some people that we need to remember everyone that AI is created for the human and to serve the human. So the first thing I should bridge and not reduce the gaps. The benefits of AI should not be limited to a small number of the countries and the companies with access to the larger computational resources. Equal opportunities need to be available for everyone, and this is particularly important for the ambitious development and the developing countries. Small states need to unite and not to be underprivileged for the languages, which is very unique, and I'm very happy that Mr. President also mentioned today the culture aspect and the focus on this. We can't forget that the human right is important, and that's why Armenia joined the coalition and on a different level, and also the Council of Europe's framework on conventional on AI human rights. This is an important milestone as well to remember that how we need to work and where to focus on. I would like— I don't want to take the whole 3 minutes, to be honest, and not to continue what was prepared to speak, and to share with all of you that we listen very often the subject that we need to unite, we need to cooperate, but something is missing, and the missing part is cooperate on what? We need to be very specific on where we want to go and what we want to achieve. So far, it seems that we have a lot of things in common and the alarm or the will is there, but the actions are a little bit delayed, and this is the place where We talk about technology, it's a benefit for technology to be fast and react rather than what we are doing as humankind. You know, it's not easy not to talk with the emotions about the technology, especially with this kind of technology. And it is, I think, very fine to show the emotions around this. And this is what the technology and the AI will learn soon. And we as policymakers and the public managers are in a very challenging situation. where we don't want to overregulate, not to harm the developing of the technology, but from other side, we also need to facilitate, not to have the fear for using that technology, but to send it on the right track. I think this is the only direction, and this is good that we are right now here, but I want to just call to keep cooperating, as it is mentioned many times, but on a very specific and precise direction. Thank you. Thank you. Thank you. So we'll proceed with the next

intervention after a small panel. I would like to call upon Sally from UNEP to introduce the panel here. Thank you. All right, thank you very much to our first

panel and first set of speakers. And moving swiftly on to our second panel of the day, and it is now my pleasure to invite Ms. Caitlin Craft, founder and CEO of Women at the Table, to introduce our second panelists. Thank you, Thank you. Hi. So now it is my great honor to introduce our panel

in order, starting with His Excellency Guilherme Patriota, Chair of the Working Group on Data Governance, UNCTAD, and Ambassador and Permanent Representative of Brazil to the WTO and other economic organizations in Geneva. Next, Deema Al Yahya, Secretary-General, Digital Cooperation Organization. Please come up to the stage and find your place here with us. Philip Tigo, Special Envoy on Technology, Kenya. Kitty van der Heijden, ASG— that means Assistant Secretary General and Executive— Deputy Executive Director, UNICEF. Jian Wang, founder, Alibaba Cloud. And Bilal Mateen from the AI panel. Thank you. You see that here? Help? Okay. So— okay, so where do I sit? Where is your— let's see. I don't have my name. We don't. Do we have a name for— so would you like to take your place then next to the esteemed ambassador over there and we'll get you a nameplate? Hi. So that's— we're getting your nameplates. Okay. As we get organized, look, it's— we have the last panel. And it's a very long day, so we're going to make you a promise that we're going to have no more principles. Since 8:30 this morning, we've heard why AI must be governed. This panel is about the machinery that makes governing possible. Because we cannot govern what we cannot measure. We cannot measure what we cannot access, and we cannot scale, what does not interoperate. That's our whole agenda: measurement, oversight, and interoperability. But one thing before we start: half of what matters is still not being measured at all. Most of the systems we will discuss today have never been evaluated on data disaggregated by sex. Keep that in mind when anyone, including me, says the word evidence. So in the most— and not in this order, so we're all going to have to keep on our toes. So to the most concrete place possible, we will begin. Energy and water. You built one of the world's largest clouds. What would honest comparable measurement of AI's energy and water footprint actually require, and what parts of disclosure should be standardized internationally rather than left to each company? Thank you, sir. Thank you, and it's really a good question.

And if you read the documents, report, just released this morning by the scientific panel, and actually there's a very informative diagram on page number 20. It really shows you the diagram how the whole infrastructure works. Okay. That one specifically mentions that relate to the energy and water. That's really the model, that's the compute, and you know that actually the AI runs on that. It takes a lot of energy and electricity to train the model. And it also takes a lot of energy to use the model. And to be honest with you, there's no standardized measurement how much it costs and what's the best way to use that. Given the scale, and I think it's a good time for us to think about that. But the one thing I want to mention is really about open source model. So when talking about open source, is very different from the open source in the software era. So today, when talking about open source model, that's basically— there's energy and electricity behind that. And so when we're talking about open source model, it's really about the open resource that you use to train the model. And so it is a new challenge that we have. And certainly, it's It's way expensive for us to use AI today, at least from our point of view. And so, it's a good time to having a kind of standardization, as you mentioned that. And the one thing that I think really need to standardize is everybody probably familiar with that. When they're using the AI application, they charge you with the token. And now, the different organization and different company have a different way to define what is the token. Okay, but that's a very, very basic measurement in the era of the AI. So it would be good that actually we're having a common understanding what the token is and make the token as cheap as a piece of paper. So today, you know, whenever grab the paper, everyone understand what it is. But today when you deal with the token and And nobody knows what it really means. And so you've got a very expensive token, and I don't think it's good for the development of the AI applications. So again, and eventually I would say that with all the effort and technology advancement and AI, when people use AI, you just use a piece of paper and a pen. Thank you. Thank you. So, from Ambassador Tigo, from measurement inside

a company to agreement among 193 states, you championed the first UN resolution on the environmental sustainability of AI at UNEA 7 and helped negotiate the very compact that created this dialogue. So this is your question by right. What do you— what would it take to get the world to agree that AI's environmental footprint must be governed at all? And what has to happen next so that resolution changes how and where compute actually gets built? Thank you. Thank you so much. And that's a PTSD question. Yeah, it

was a difficult to negotiate this. And maybe I'll start from there. So, I think, and I agree with Jamila's point from the previous panel around our framing around AI and its impacts on environment. And so, I want to frame it based on our thinking from, of course, negotiating the Safe Sustainable AI Resolution, but also the one for UNEP. I think 2 ideas, of course, AI for Green, but also Green AI, right? So, I think it has to be about opportunity for countries like mine. that could leverage the transformative power of AI, whether it's to improve food systems and food outcomes, that actually means better income for farmers, but also it means a farmer who cannot afford a crop cycle to fail, and which is a direct route to poverty, has a means to survive. So it has to be that practical, that the idea of the human in the loop is just not about in the technology itself, but in the design that we must place humans at the at the center of designing artificial intelligence. I think, secondly, I think it's how do we also practically include environmental sustainability in the COP process, in the nationally determined contributions, because I do not see much of AI in NDCs. I think third is green AI, right? So the second part, which means AI needs to become part of environmental toolkit, and its footprint becomes, I think, part of governance. And I think people talk about governance, but not necessarily necessarily from mine to model. So we govern the impact of the model and not necessarily from the mine, rather the model, which means water, minerals, land, compute, children working in mines, which is my third point around safety, that we must broaden safety definitions, not just technical safety, but also socio-technical risks, which includes environmental environmental risks, which means human in the loop, again means humanity first. So I think for me, as I end, a couple of things. One, and the scientific report kind of mentioned this, that we must recognize environmental resources implications, but also I think there's a gap there, and maybe this is what the dialogue must talk about, and I know Jeanne has mentioned this, that when you talk about standards, but also I think for me, standard measurement means in the reporting across the AI lifecycle, which means we must have the confidence right now in something that the UN High-Level Advisory Body, which I was part of, mentioned around a global environmental AI standards exchange. Because, yes, you talk about— everybody knows about paper, but paper in my country is different from paper in— So we need a standards exchange, which means we need a common understanding. But also, we need to think about measurement and governance not from a punitive perspective, but from a capacity-building perspective, because once you talk about punishment, people tend to be less around transparency and accountability. And I think as I end, let's think about from where we start, right? So I think something about artificial intelligence, but also as we measure, it is new, and potentially it is very scary for people. But we must think about environmental transition at the same breadth as AI transition. And so we should not be looking at neat solutions. We should be very much looking at how these two are interconnected, but at the same time, we must realize that at the next frontier of AI must be sustainable AI. It must be by design. And with that, I end. Thank you very much. Thank you. So you talk about impact, we talk about standards now

a lot, and we turn now to Kitty van der Heijden of UNICEF. And I— my question to you is, a third of internet users are children, and almost none of the design, to put it mildly, the choices are shaping their attention and decisions were made with them in mind. What concrete standards or audit requirements would actually protect children from manipulative design, and who should be accountable for enforcing them? Thank you so much. And there's no question that could be more relevant.

I know women are half of the world, but children are half of the world's extreme poor. And if we do not protect them, all of humanity will suffer. That's what we at UNICEF work for. And it's very clear that children are the most exposed to AI and should be the ones that we protect most. And we have new research published in recent days in 10 countries which clearly shows that children are adopting AI 3 times as fast the adults that are raising them. And not only that, AI is woven into every system, every service that these children are accessing in a very invisible way— what type of information they get, how the system will treat them. And on the one hand, and we talk about that a lot, obviously, AI can be the most, the fastest accelerator for humanity. Look at what they can do with personalized training, with translation of languages, Think of assistive technology. Think of health information reaching the most remote communities. But let's not forget that it is only an opportunity multiplier because if we build it like that, and on today's trajectory, how it's built, it is built for certain language groups, for children that are connected from relatively affluent societies, and for every other child, what this is going to do, it's going to entrench the learning crisis, and it's going to automate exclusion. That is not what we should be wanting to do. And so that doesn't mean, which is the flip side of the argument that I keep hearing, oh, you want to slow down innovation and opportunity. Of course not. We need to accelerate it. At the same time, we need to steer it, just like we steer and regulate the food that we eat, just like we regulate the medication that we take, because we know that children will be exposed exposed to that. And AI should be no different than that. Good governance is how we scale the positive impact and avoid scaling the harm. It's as simple as that. So how do we do that? Making this an opportunity multiplier for every child? I think 3 key things we need to do. One, design for inclusion from the start. Not as an afterthought, not as an add-on. Oh, God, the algorithm did it. No. You build a system with children from the get-go. Otherwise, you're building the exclusion engine by quietly misqualifying, denying children that are not part of the datasets that we're currently feeding into AI— children with disabilities, children from the South, children in crisis settings. If we exclude them, we are not going to build the system that we need to see. Second, we need to treat children's data not as raw material, but as a rights issue. Children now have a data profile basically from the moment they are born. It's built out before they can even spell their own name. By the time they are 13, they can get a data profile till they are in their grave. Those data belong to children. The assets that this generates belong to children. We need child data to be used with child safeguards. And third, if we use AI in schools, in clinics, they have to follow a child development logic, not a commercial logic. A school, it needs to be very clear, is not a sales channel. A clinic is not a marketing opportunity. AI can never be a substitute for real learning, for real teachers, for real care. As the co-chair Of the dialogue said at the press conference just now, I do think that children's rights are one of the few things that all the member states can probably rally around, can agree on. So let's use this conversation to make sure that the children, as the Secretary-General said this morning, are not treated as guinea pigs. We owe them much better than that. Thank you. Thank you so much, Madam van der Heijden. So we're talking about building,

about inclusive, robust design, something that's transformative, that brings benefit. I'm going to go now to our colleague from the scientific panel, who is here as our honest broker. What does the evidence actually tell us about which oversight and transparency mechanisms work, and where's the science too thin for governments to act, whether because independent researchers cannot get access to the systems or because the data was never disaggregated by sex, among other things, to show who a system actually works for and who it quietly fails. Thank you. Thank you so much. We've talked a lot about environmental implications,

so I want to anchor us in a historical metaphor that I hope many of you will find familiar. So we're in 2015, we're sat in Paris, we're talking about negotiating a treaty on climate change, what we should do after 25 years of work, 5 systematic assessment reports, decades of coordination. And just half a decade later, the same UN Secretary-General that stood in front of you this morning stood in Glasgow and said that the 1.5-degree target, that threshold that we needed to stay under to prevent catastrophic risk and impact on human health and well-being, was on life support, was on life support, and the machines were rattling. We're not sat in Paris, we're sat in Geneva. And in the last few years, we've gone from the advent of transformer architecture to over a billion users of singular technologies to profoundly concerning capabilities of those same AI tools. We may well be reliving history, but on a hugely compressed timeline. We won't have to wait another decade to see both misuse and the impact of missed opportunity due to a growing digital divide. In short, there are definitely places where the evidence is thin, right? There are also places where the evidence is robust. The one thing that I would like everyone to take away is that delay on either of those topics is nothing short of a dereliction of duty, right? The question is what we do for each. Where the evidence is, I think, reasonably robust to justify action, where it concerns sexual violence against women, against children, right? Companies institution model developers must be held to account. We know that these general-purpose technologies can be misused in that way, and action must be taken. Similarly, we've seen the impacts of a lack of interoperability in the digital health space. It undermined our response to the Ebola crisis in West Africa over a decade ago. It's doing so again. We're going to see the same implications of a lack of interoperability standards for AI, right? It's going to leave the global majority out of the conversation. Now, to that point about where the evidence is thin, it's the nuance in what we do, right? Standards aren't just about setting the bar for what good looks like. Standards are about building transparency into the system and making sure that we're all reporting metrics that we can compare. There is huge work for the international community to invest in to build those transparency-enabling standards. And the last thing I'll leave you with is I'm a physician by training. I work on the regulation of AI as a medical device, and the tension that I constantly hold when I'm asked about standards is radical transformation versus being rational. We have platforms that work, that help us do oversight of medical devices. Equally, those same systems have been challenged by generative AI. My question is, how do we come to those same systems? How do we abandon the vestigial institutions we no longer need but retain those elements that have been so useful. Thank you. Thank you very much. I like the phrasing radical transformation versus the rational.

And that brings us to our next colleague, who's going to take us from national practice to cooperation between unequal partners in DCO. DCO's membership spans very different levels of AI maturity. Yet you're building, Madame, shared frameworks across all of them. What actually makes interoperability work between companies— countries with unequal capacity? And we'd love one example of something that worked and one that you would tell this dialogue not to do. Thank you. Well, thank you so much for that question. Actually, we— The essence of the creation

of DCO as a digital cooperation organization is to create a platform to convene and to cooperate and to share best practices together as nations. We truly believe that every country has a competitive advantage. Identifying that competitive advantage and linking these competitive advantages together, they create a big force in terms of bridging that digital divide and also narrowing the divide that we're seeing from AI. And I'll give an example. Does every country need to invest and put capital in a 100-megawatt data center, for instance? So first, identifying what— where are you from the value chain? What can you add value in that value chain? And then you contribute and create the right frameworks and bankable projects with the private sector to create an accelerate— that acceleration to growth and prosperity in digital economy. For instance, countries that invested in compute infrastructure, we have other countries that have the brainpower. Creating a framework and a governance that, for sharing IP and sharing best practices that gives opportunity for countries to create local content faster till they have developed their digital infrastructure. And on another hand as well, having the private sector in the center of the discussion, having the innovators be part of the co-creation and co-design of policies, regulations, as well as initiatives that supports in accelerating the— and narrowing the gap of the digital divide, especially in countries in the Global South, as well as LDCs, who not only need a basic AI infrastructure, they need basic connectivity. Thank you. And now, Ambassador Patriota, you're chair of the CSTD Working Group on Data Governance, and

you close this arc with the question no one else in the building can answer. All day this room has discussed governing AI models. Your working group is drafting its report to the General Assembly right now, and one of its tracks is interoperability between national, regional, in international data systems? Where is real agreement emerging? And what is the one thing this dialogue should say about data, given that no commitment on AI means very much without it? Thank you. Thank you very much. I think that's the first thing they should say, that data is a foundation

upon which AI governance will be based. So the governance of data, I think it's at the bottom, at the baseline of this pyramid. And that's what we're trying to tackle with in the Working Group on Data Governance. The Working Group to me is a very interesting new experience within the UN system because it is a hybrid entity. It has half state representatives and the other half— 27 members each— the other half is our non-state participants. And incredibly, we have managed to have a very constructive dialogue within the group. We've had 6 sessions. We are at the point where we have actually a zero draft, which is quite an achievement considering it's the first time anything has been attempted regarding a text within the UN limits of words. which has— means it has to be short, like the 3 minutes here, that people get cut off when they overstep the 3 minutes. And we have members of our group participating actively here today. Nick Ashton-Hart is one of them. We had Anja Caspersen, who spoke as well, and many others are present. So, very representative group. Interoperability is one of 4 tracks. We have the foundational principles as such. We're trying to figure out what principles would be interesting to highlight with respect to data governance. Some are extracted from existing platforms and agreements and even national experiences. We have another track on interoperability, as exactly that, your question. We have another one on sharing the benefits of data, which is a very sort of development-oriented north-south or digital gap kind of discussion, and then we have cross-border data flows, how to make it flow. The cross-border data flows interacts with the interoperability. They could have been just one subject, but we decided to split into 2 to give more balance and more focus. On the interoperability front, we don't have an agreement yet. We're working very well, but I think the first highlights to make is that interoperability is incredibly important It's desired. It is desired from both a development perspective, but also from a business and entrepreneurial perspective. We cannot choose one over the other. I think that's one of the first conclusions. And it doesn't necessarily mean harmonization of national jurisdictions or laws or regulations. I think we can achieve interoperability without that, in fact, by— through international cooperation, through agreements, through different contracts, in fact, common understanding. So that's, I think, an interesting thing because people tend to see a tension between the two, an opposition between interoperability and, in a way, the sovereignty of laws in the national space. And that doesn't have to be like that. I think we can, through international cooperation, we can find solutions that will safeguard national sovereignty and also promote interoperability. And data flows. And all that respecting human rights. It's a very big issue that has appeared up on the priority list, especially children's rights, which was, I think, the focus of Kitty's intervention and the UNSGs this morning. I was— I found it interesting that when he managed to mention killer robots, everybody clapped, but when he spoke at length about children, people did not clap. They should have clapped. So that's my— Small contribution here. Yeah. And finally, I think that there's an underlying— interoperability is considered an enabler of development, but not the goal. So, and there are underlying conditions to achieve that as well, that many developing countries have— do not currently have access to. So, it's capacity, skills, financing, all sorts of things. We heard here about also the environmental impact of this, of the data centers, very costly, consumes a lot of energy, water, etc. So all these things are developmental issues and the focus of everything we do in the group is as relevant for development because it is a United Nations group. And hopefully we'll have a report ready for you by the end of the year. We're going to send it in April to the Commission on Science and Technology for Development, and from there it should go to the General Assembly at its 81st session. Thanks. Wonderful. So now we're going to have a lightning round to close up. As anybody who knows me knows, my panels

are fast. So we're going to start in the same order that everybody spoke, so we'll start with Mr. Jian Wang, please. I think it's very fascinating if you think about technology, but also, you know, social, economic, cultural,

all the issues are very important for the healthy development and really do good things for the people. Thank you. Thank you. And let's also just be aware that we're speaking directly to the penholders for the global dialogues

we want to have. What do we want included there, too? So I'm going to ask Ambassador Tigo. Of course, I've talked about the Global Standards Exchange, but I think also we need to think about, yes, interoperability

of systems, but also human interoperability, because I think what we are missing is the culture of working together and breaking the silos. Thanks. Madam van der Heijden. Thanks. And I come for half of the extreme force. I have 3 asks. One, we really need a global evidence baseline

on children's AI access,

how they use it, their literacy, their development impact. And that should be treated as a global good. Without that, we are essentially Running a worldwide experiment with children as the guinea pigs. Sorry, we cannot allow that to continue. Two, we just need something very simple: child-centric benchmarking. What's the metric of success which needs to be built in those models and how they are evaluated? And there must be very clear child rights red lines. We do not go there, and child sexual abuse material is one of them. Zero tolerance. And three, child mandatory. child rights impact assessment when AI is built into systems that— for services that children depend on: health, education, child protection, migration, welfare. If not, children will suffer the impact. Thank you. Thank you very much. I'd like to add, just as long as this is my only time, an integrated impact assessment that takes environmental concerns,

human rights concerns, gender, children's rights concerns, and brings them together so it's a one-stop shop and we're not forced to trade off one issue after another. Now I go to Bilal for his lightning round. Thank you. It will come as no surprise to anyone that the scientists want more research funding, but every nation-state

must make a commitment to funding foundational research in AI. It matters who does the research that determines what research is done. And for countries to have a meaningful seat at the table, they must be driving the scientific agenda on the topic. Fabulous. Madam El Yahya. Well, thank you. Yes, I would really focus on 3 main points, actually. First is the data. We really need to focus

on the quality

of data that we're feeding the AI with. Because the more the AI gets intelligent, it is going to be built on that fundamental data. Second, local content. We have to enable nations and countries to create their own models, to create their own algorithms. We can't continue becoming consumers and users. We have to become producers, and we have to increase the options, not on a national level, but on an international level. Third, last but not least, which is the resilience of the infrastructure. We have to protect the infrastructure against any environmental catastrophes, God forbid, political tensions, or even economical crises. Thank you. Thank you. And we close— thank you— and close with Ambassador Patriota. You're Thanks. My ask is to treat data very carefully

and fairly when you're discussing AI governance. AI gobbles up data, but data

belongs to people. Data reflects people's privacies, people's rights, children's rights, adults' rights, and they are not necessarily freely available, or they access to data has not necessarily been given consent. Correct. So this— we need an architecture for that. We need a decent framework for that. No one better than the UN and through the global dialogue to achieve that sort of common understanding as to how to treat data with utmost care and also to be fair and to share the benefits of data with those who have produced the data at the baseline. Data is also part of the creative industry. A lot of creators are being misappropriated. Their data has been— is being misappropriated through AI processes. I think we can achieve a decent understanding as to how that should work in practice and be fair about it. So fairness is my ask. Thank you. Thank you. And with that, we conclude our panel. Thank you very much. All right, so we'll start with the intervention session again.

I would like to call upon the stage the representative from Anglia

Ruskin University. On the stage, please. Thank you. If we don't see you, then we can move on. Okay, we'll move on. Next, I would like to call upon the Counselor to the UN Office and other international organizations, Geneva, from Slovenia. Distinguished co-chairs, dear colleagues, ladies and gentlemen, what an inspiring conference. We in Slovenia, we see AI as a general-purpose

technology whose benefits should be widely shared across societies and economies. From a social perspective, to start, AI has the potential to improve public services, healthcare, education, public administration. Many speakers have said this earlier today, but realizing these opportunities requires investment in skills, public trust, and responsible deployment of AI. Economically, AI by no doubt offers significant opportunities to enhance productivity, innovation, competitiveness, particularly for SMEs. Our focus in Slovenia is therefore on creating the conditions for wider AI adoption through infrastructure, knowledge transfer, financing, and above all, stronger cooperation between research industry, and the public sector. And as a small community, Slovenia attaches particular importance to the cultural and linguistic dimensions of AI, as Finland, we heard today. We believe that multilingualism and cultural diversity should remain strengths in the digital age. AI should support all languages, including those with smaller speaker communities, ensuring that digital transformation is inclusive and that no language is left behind. At the same time, all these opportunities depend on strong technical foundations. Access to compute, high-quality data, interoperable infrastructure, and digital skills remains essential for the responsible development and wider uptake of AI. So in this regard, Slovenia is investing national AI infrastructure. We need it, including Slovenian AI factory and AI competence center. In our review, continued international cooperation will be essential. By sharing knowledge, strengthening capacities, we can ensure— we believe that sincerely— that AI contributes to sustainable development and that its benefits are accessible to all. Thank you very much. Thank you very much. We will now invite the representative of AI for Peace, which will then be followed by the representative of the Republic

of Korea, Diversa, and Australia. AI for Peace. 1, 2, 3. Okay. Perhaps the delegate from the Republic of Korea. Thank you, co-chair. Excellencies, distinguished participants, the Republic of Korea welcomes this first Global Dialogue on AI Governance.

In this spirit, the Republic of Korea offers 3 observations on AI governance. First, the gap between the speed of AI development and our readiness. Is becoming apparent. AI is advancing faster not only than our laws and institutions can follow, but also faster than our society can be adapted to it. The Republic of Korea has felt this firsthand. When we brought AI tools into our classroom, we realized that the gap between rollout and readiness and found the different takeaways among various stakeholders of AI classroom. The same pattern is visible well beyond our borders, from self-driving vehicles to the spread of misleading AI-generated content. Young people are increasingly worried about the implication of AI on their job opportunities. These instances emphasize the urgency of broader discussion on how to build a social consensus around those challenges at a speed commensurate with the— at the speed of AI development. Second, we are witnessing not only harms already known but also harms previously unknown. Just as the potential application of AI continues to can even in unthinkable ways and areas. The range of complexity of risk it presents may be beyond what we see now, such as psychopathy. Here we must be ready to respond to the risks and harms have yet to emerge. Third, the distance between our respective endeavors. As countries deepen in their led in this for AI, so does, so does their approach to how to govern it. And globally, these efforts remain fragmented. This is why the Republic of Korea declared our vision for establishing global AI hub together with 9 participating UN and related organizations to contribute to strengthening of global AI capacity. In conclusion, we must keep sight of what all of this is for. The promise that humanity sees in AI is not in how far the technology advances. It is how the technology serves people. The Republic of Korea stands ready to do it, to do its part. I thank you. Thank you. Thank you to the Permanent Representative of Republic of Korea. Next, I would like to call upon the Executive Director of Diversa onto

the stage, please. Thank you. Thank you. Thank you. Thank you so much. My name is Diana I'm Diana Mosquera. I come from Ecuador. 6 years ago, I co-founded Diversa, where I serve

as executive director now. And we develop AI data and technology from a critical perspective, centering on people, human rights, nature, and social justice. Okay. We often talk about AI as a tool to accelerate sustainable development. But we talk much less about the infrastructure that makes it possible. As the Secretary-General mentioned this morning, it is not credible that AI will help us respond to the climate crisis if we do not identify and seriously address its environmental impacts across its full life cycle. Every AI model exists in a very material way. Independence of data centers, large volumes of water, intensive energy, critical minerals, global supply chains, and human labor. And all of this remains almost invisible in most governance debates. If we want AI to truly contribute to sustainable development, we need to govern not only the algorithms and the AI final products, but also the infrastructure they are built on. Governance here is not only technical or regulatory, it's political. Deciding what gets measured, disclosed, and audited is itself a form of power. From that perspective, I want to raise 3 priorities. The first is to make environmental transparency a technical principle of AI governance. Today, there are still huge gaps in measure and disclosing in a comparative ways the energy use, weather footprint, critical materials, and e-waste linked to the lifecycle of AI. The second priority is to build technical capacity in countries, especially in the Global South. Governance cannot depend only to who develop the models. Countries need the capacity to evaluate, audit, and adapt adapt AI systems according to their own needs, social, environmental, and regulatory contexts. And the 3rd priority is to recognize that AI infrastructure has territorial impacts. Communities where the minerals are extracted, where data centers are built, and where environmental costs are concentrated must be part of the decisions about how these technologies are developed and deployed. To conclude, we should not take the development of AI something already decided or a path that we simply have to follow. We can still choose what type of AI we build or for whom. That means making visible everything that involves development of AI and moving forward together, together in a technical responsibility. Thank you. Thank you. May I please invite the Ambassador for Cyber Affairs and Critical Technology of Australia, Your Excellency, on the floor, please. Thank

you, co-chairs and distinguished delegates. Australia welcomes the discussion on the AI opportunities and the implications so we can accelerate

the benefits and ensure that AI is safe, secure, and trustworthy. We know that AI provides unprecedented opportunities for sustainability and inclusive development, including by improving productivity, contributing to disaster risk reduction, healthcare, and climate change mitigation applications. We are also aware of the risks that irresponsible use of AI presents. These emerging challenges demand nuanced and flexible solutions that strike the balance between supporting innovation within secure means. As outlined in Australia's National AI Plan, our focus is straightforward. First, to make sure that we fully capture the AI opportunity by building the systems, infrastructure, and frameworks to enable confidence in the take-up and use of AI. Second, that we share the benefits with all. We know that AI can widen existing disparities across the ecosystem, and so we need systems and frameworks in place that are interoperable to ensure that these benefits reach near and far. And lastly, we support the safe and secure use of AI. That is why Australia has established the National AI Safety Institute to strengthen strengthen Australia's ability to understand emerging AI capabilities and risks, and to support evidence-based responses as technology develops. And Australia is engaging in international forums like this with full cognizance and awareness of the challenges and opportunities of our time. AI governance should ensure these opportunities are realized in ways that are safe, inclusive, and consistent with international human rights law, while addressing privacy, equality, freedom of expression, cultural and linguistic diversity, and meaningful human oversight. And we are better placed to do this together than alone. We need to also pay close consideration to cohorts already disadvantaged by digital and economic gaps, as well as those at higher risk of AI and automation-driven disruption. This includes our First Nations people, women, people with disabilities, and regional and remote communities. And this is why Australia has committed to empowering our First Nations people through aligning our action on AI to our national agreement on closing the gap and working in genuine partnership with First Nations community. The message from the UN independent international scientific panel is clear: that success in AI is not solely dependent on who has the most chips, data centers, or the most capital. Success is achieved when companies— countries adopt technology which is backed by responsible governance, broad and equitable access. AI's greatest benefits will not come from technology alone, but from the choices we make to govern. Thank you. Thank you. I would like to invite the stakeholder from NASCOM AI Please. Okay, it will be followed by His Excellency the Ambassador of Mexico. Thank you. Talking

about multilingualism, hablaré en español. I'll take the floor in Spanish, says the speaker. Very good afternoon, distinguished delegates, representatives,

co-chairs. METSCO is of the view that it's important to address in better detail technology such as AI and what it can do with human beings, and not to limit ourselves to combat what AI can do for us. Thus, we welcome the preliminary report from the independent international scientific panel's report. In particular, this is useful when we look at the outcomes from the working groups from 5, 6, and 7 pertaining to, inter alia, human rights, democracy, and cultural diversity. The rapid development of AI is already changing the social contract, redefining power levels, economic activity, and social networks. With this, it's absolutely critical to have open dialogue that is inclusive between governments and representatives from different stakeholders. International cooperation, as has been reiterated here a number of times today, is absolutely paramount to address the challenges before us all. Mexico has included technological development as one of the cross-cutting pillars of its National Development Strategy for 2024 to 2030. Against this backdrop, and under the guidance of the National AI Development Agency, We'll be able to look at development and transformation as a driver for human rights. We're trying to find technological solutions which are public, thereby consolidating common technology as a common right. Mexico is of the view to look at the different identities which are both cultural in order to avoid technology development which is only based upon inequality. In this vein, We'd like to analyze new ways to look at integration of these different identities and to continue to develop language models. Perhaps we should review and consider once again concepts, new concepts as well. To close, in order to adopt the new realities for different countries such as those in the Latin American and the Caribbean regions, it is necessary to undertake an analysis of the, of the impact on the sector of AI. Thus, Mexico, in cooperation with the program, the UN Development Program, is going to convene a regional forum at high level at the end of this year in order to address this very topic. The outcome of this will seek to have, uh, to ensure that Mexico cooperates in the global dialogue on artificial intelligence. With that, I thank you. Thank you. I would like to call upon the Principal Officer of Dubai Municipality onto the stage, please. Okay, moving on. Uh, I'd like to invite His Excellency, Special Envoy of the President

of Sri Lanka, onto the stage, please. Co-chairs, Excellencies, distinguished colleagues. As AI increasingly becomes the interface between people and knowledge, public services and economic opportunity, inclusion will

no longer be determined simply by whether people have access to AI. Inclusion will depend on whether AI understands the people it is intended to serve. The first dimension of this challenge is language. Today, several hundred languages and dialects remain underserved in terms of limited datasets, language technologies, and AI models. If left solely to market forces, AI will naturally evolve around the world's largest languages, leaving many communities on the margins of the AI economy. Closing this gap requires investment in language equalizing technology, linguistic datasets, and foundation models. But language parity alone will not suffice. True symmetry also requires culturally inclusive AI. An AI system that understands my language and dialect perfectly but does not understand my cultural context is only partially inclusive. An AI system that understands but cannot recognize local value systems, local value systems, traditions, institutions, farming practices, legal systems, or cultural norms, it cannot provide advice that is trusted or relevant. AI must understand not only how we speak, but also how we live within our societal construct. Deep and meaningful inclusion then raises the imperative of sovereign AI, the ability for countries to deploy AI in ways that reflect their own national priorities, cultures, and languages, without importantly having to compromise independence, sensitive data, or national resilience. To reach these ideals, Sri Lanka's digital transformation architecture places AI-powered language equalizers as digital public infrastructures. Language models, speech recognition, optical character recognition, and translation services will become horizontally platform capabilities. Simultaneously, AI will also enrich adjacent DPIs such as conversational government information services, and multilingual farmer advisory services. Investments in these capabilities are investments in equal opportunity. But unless international cooperation bridges both the sovereign AI gap and gaps in localization, some communities and countries will be left behind. Smaller and developing nations must not be constrained trained to being just consumers of AI, but be empowered contributors to the knowledge, language, and cultural perspectives upon which future AI systems are built. Ultimately, the success of AI will not be measured only by the intelligence of its models. It will be measured by the diversity of the people they serve. Thank you. Thank you very much, Your Excellency. We will name the next 4 people. First, the stakeholder from Empathetic AI, the founder and CEO. Then His Excellency, the permanent representative from

Belarus. Then the WWF International and Côte d'Ivoire, and my colleague Kuchia will take over. Thank you so much. Empathetic AI? No? Okay, so then, um, the permanent representative to the United Nations office from Belarus. Thank you, Your Excellency. I'll be continuing the tradition of multilingualism in this room. First of all, sincere thanks to all the participants. Belarus is a middle-income country and therefore

can't take part in a competitive race for budget and quantity. However, we have good human resources. Done good work on digitalization. And for us, AI is a tool to improve the quality of life of citizens, to modernize our economy and various spheres of public administration. We are working on a system to digitalize our public administration services by analyzing big data and ensuring preventing services. We are deploying technologies in healthcare, education, and the social sphere. The idea is for new technologies to ensure better access to better services for every citizen wherever that person might live. We're also seeking to optimize our industrial and agricultural processes. We are deploying solutions for logistics and urban planning, and we also have targeted solutions for climate monitoring, for managing natural resources, and for protecting the population. The deployment of AI does not just change the economy and the social sphere, not just the labor market, but also the environment in which we live, in which we work. Therefore, the issue of human control over critical decisions remains extremely relevant. Our responsibility for the decisions of the algorithms, the protection of data, and the prevention of discrimination. Human beings are at the heart of the technological revolution, and we need to focus on human behavior first of all. There are 2 fundamental issues that we need to find We believe that the global dialogue can become an important platform in order to find the answer to these questions and other important questions. AI must develop for the good of humanity, and it must be based on the accountability of developers, the protection of human rights and human dignity. Thank you. Thank you, Her Excellency, for your time. We'd also like to request, uh, we would be extending the session by 15 minutes, so would be grateful if all of you can be seated. And as we go through the session to complete

the interception followed by the closing remarks, the interpreters can stay on. The interpreters can stay on. stay on till that time. Yeah, thank you. I'd like to call upon the delegate from WWF International onto the stage, please. Thank you. Thank you. Moving forward, I'd like to call upon the minister from Côte d'Ivoire onto the stage, please. Allow me to speak. speak in French. Ladies and gentlemen, dear delegates, as many speakers have mentioned already this morning, AI requires considerable investment and cross-cutting coordination for those

countries who have cultural, linguistic, and sociological sociological dimensions. We must ensure we don't repeat the mistakes of the past. In the area of telecommunications, for example, at the beginning of industry, we've seen fragmentation of regulation stemming from— which has led to a a delay in the development of markets. But the community framework today and the common sector policies which see a synergy framework in order to build a common ecosystem to develop AI— Côte d'Ivoire and other countries from, from the West African region have always been involved in the community participation in order to develop, for example, something that's been set up by UADA and by— towards economic and monetary integration, which has been set up over a month ago. These institutions have offered fertile ground for AI which is adapted to our needs. This is a lever for development which should be exploited from today. Therefore, I'd like to launch an appeal for regional cooperation pooling of efforts for projects and investment to build infrastructure together for management tools, for our informational heritage in order to build AI for all, to build educational systems, and for our financial markets. We must undertake efforts to harmonize our systems in order to lead to more prosperity by working on our common values. Therefore, we can bear the, the rich and linguistic richness so that AI can from today benefit all our people in the areas of health, security, and education. Côte d'Ivoire has a long tradition of hospitality given its strong commitment to international institutions such as the such as ITU. Côte d'Ivoire would like to solemnly appeal to the countries in the West African region to work together. The interpreter apologizes, the speaker's microphone cut out. Thank you. Thank you very much. Your Excellency. So before I proceed, I would like to ask for the indulgence of the interpreters. We are running 15 minutes late. Would love to seek your indulgence. Chair, we can provide interpretation

until 6:15. Thank you for your understanding. Thank you so much. Thank you. Would like to invite the AI Research Fellow from the Council on Foreign Relations. Good afternoon, co-chairs, Excellencies,

distinguished delegates, ladies and gentlemen.

It is a pleasure to have the opportunity to intervene today. This is a particular pleasure to address this dialogue, having had the honor of representing

134 developing countries in the negotiations to establish both this body and the Independent Scientific Panel on AI. In my current role at the Council on Foreign Relations and as part of the Lead AI team, we work from a single premise: AI deployment is foreign policy. It's the least governed and among the most consequential. The governance architecture assembled over the past 5 years is oriented overwhelmingly towards the moment before release. It asks whether a system is safe to deploy. It far more rarely asks whether the people that system later reaches in a clinic, a benefits office, or a classroom retain any means to contest, correct, or seek remedy for the decisions made about them. That is the gap. Traceability without recourse is accountability in name only. The evidence is not speculative. When an automated system flags citizens for fraud or denies them benefits at scale and no identifiable decision maker stands behind each determination, harm compounds for years before any courts intervenes. A model trained under one jurisdiction and deployed into another arrives without obligations, oversight, or avenues of redress that any affected population would expect of an authority acting upon them. My submission on accountable design is therefore this: standards will be effective only if they are operationalized— if they only— only if they operationalize 3 requirements at the point of deployment. First, accountability, clear attribution of responsibility, across the supply chain so that a deploying entity, not an opaque system, is answerable to the jurisdiction it affects. Second, due process, formal pathways through which affected people can appeal an adverse outcome and obtain remedy. Democratic institutions have spent centuries building such mechanisms in other domains. Our task is to adapt them, not to invent them. Third, openness, mandatory disclosure that makes deployed systems legible to the communities and authorities they touch, backed by guaranteed access for independent researchers, without which oversight is a claim rather than a practice. For the co-chair summary, I offer one priority for continuity into 2027: that this dialogue treat governance at the point of deployment as a distinct work stream so that accountability, due process, Thank you very much. UN-Habitat. Yeah, we'd like to invite the delegate from UN-Habitat. We'll move forward. We'd like to invite Her Excellency, Minister of Digital Economy and Transformation of Togo, onto the stage, please. Moving forward, I'd like to invite the delegate from

Guatemala onto the stage, please. Guatemala, maybe here Guatemala. Okay, they come. Dear moderators, distinguished delegates, artificial intelligence represents an opportunity to promote sustainable development. Sorry, I will change to Spanish. Sorry, I was running. Guatemala. Guatemala is of the view that AI is an opportunity

to drive forward sustainable development for our country. Ethics is a cross-cutting principle that should be geared towards all stages of the cycle. The lifecycle of AI, from its design development towards its implementation and usage and its assessment. This approach needs responsibility to drive forward trust from the public and will contribute to prevent risks associated with to bias, exclusion, multilingualism, and multi-ethnicity. Guatemala therefore is of the view that it is extremely important to consider the cultural and linguistic diversity of AI. We believe that diversity of knowledge, language, and experience will broaden perspectives from which AI solutions can be found that are more relevant and will be more pertinent to the different linguistic and cultural contexts. Guatemala reaffirms its commitment to continue promoting AI that will harmonize innovation with ethical responsibility, will respect human rights and will value linguistic and cultural diversity, which will strengthen national capacities and will contribute to sustainable development and to the well-being of our societies. Thus, Guatemala will continue to actively participate in the international cooperation fora in order to share experiences, to craft consensus, and to promote AI governance that is even more inclusive, more representative, and human-centered. With that, I thank you. Thank you. I would like to invite the delegates from the Philippines, please. Good afternoon, Chairperson, Excellencies, distinguished delegates, ladies and gentlemen. The Philippines welcomes this opportunity to participate in the first Global

Dialogue on Artificial Intelligence Governance. We extend our deep appreciation

to the United Nations for convening member states on a matter that now bears directly on economic participation, public services, culture, language, education, employment, and the exercise of rights. For the Philippines, the implication of AI must be understood through the different ways people encounter technology. The Philippines recognizes the opportunities that AI can offer. It can also strengthen public service delivery, support disaster preparedness, improve access to education and health, assist farmers and enterprises, and expand the reach of Filipino creativity. At the same time, the risks are very real. AI can displace tasks, widen skill gaps, reproduce bias, enable manipulation, misuse personal data, weaken accountability, and place cultural and creative expressions at the risk of unauthorized use. These risks do not fall evenly across societies. Workers, children, and persons with disabilities, Indigenous peoples, creators, small enterprises, and communities with limited digital representation may face particular exposure. For this reason, the Philippines believes that AI governance must remain centered on people. In the world of work, Technological change must be accompanied by practical pathways for adoption. Workers need access to lifelong learning, reskilling, upskilling, and credible transitions to new forms of employment. Enterprises, especially micro, small, and medium enterprises, require support to adopt AI responsibly and productively. In the creative sector, governance must address recognition, consent, attribution, and fair remuneration. Creative works, cultural expressions, and traditional knowledge shared not be— should not be treated merely as raw material for technological development. In education, health, and public service, safeguards must be very clear. AI systems used in sensitive settings should be subject to appropriate oversight, transparency, testing, and mechanisms for redress. The Philippines supports international cooperation on inclusive AI impact assessments, culturally and linguistically representative datasets, provenance and attribution standards, accessible redress mechanisms, and stronger protection against discrimination, manipulation, and harmful automation. We also emphasize the need for developing countries to participate in shaping these approaches. Countries should not only receive AI systems delivered elsewhere, they must also be able to build, adopt, evaluate, and govern systems that reflect their own languages, institutions, cultural context, and development priorities. Lastly, the Philippines remains committed to constructive international cooperation in shaping AI governance that supports innovation, protects rights and respects culture, as well as expands opportunity. Thank you very much for your attention. Thank you very much. We'd like to invite the distinguished delegate from Egypt. Thank you. Thank you. Thank you. Co-chair, Excellencies, ladies and gentlemen, Egypt welcomes this opportunity to focus on the operational dimensions of AI governance. This breakout

session is where our shared ambitions must move beyond principles and translate

into practical action. For the Global South in general, and our continent, Africa, The success of AI governance will ultimately be measured by whether AI becomes affordable, accessible, locally relevant, and trusted by our people. With that in mind, Egypt would like to highlight 3 priorities for the co-chair summary. First, we must redefine what meaningful access to AI means, bridging the AI divide, requires more than capacity building. It requires investment in compute infrastructure, sovereign data capabilities, and sustainable financing mechanisms that enable developing countries to build their own AI ecosystem. At the same time, AI development must be environmentally sustainable through the promotion of green AI and resource efficient models, frugal AI. Second, we must prepare our public institutions and societies for AI capacity building. And for AI capacity building should extend across the entire public sector, while reskilling initiatives and just transition policies help ensure that AI enhances human capabilities and that its economic benefits are shared equitably. Egypt pays very special attention to AI literacy for children and minors. Third, AI governance must reflect cultural diversity and prevent excessive technological concentration. Developing countries must be active contributors to AI datasets, standards, and models while global interoperability frameworks must incorporate cultural, linguistic, and socioeconomic diversity. Ladies and gentlemen, to translate these priorities into action, Egypt proposes 4 practical outcomes: expanding equitable access to compute infrastructure through international support mechanisms, promoting global repositories of lightweight AI models, developing practical AI procurement guidelines for the public sector, and embedding cultural and linguistic benchmarks within global interoperability and AI safety frameworks. Ultimately, meaningful inclusion in AI governance is measured not by participation alone, but by influence. Egypt therefore calls on the international community to ensure that equitable— thank you. Thank you. Next, we'd like to call upon the Ambassador to the WTO from Costa Rica onto the stage, please. Thank you. Excellencies, distinguished delegates, artificial intelligence is often described as universal opportunity, but opportunity is not universal when computing

power, investment standards, languages, and decision-making authority remains concentrated

in a small number of countries and companies. Latin America and the Caribbean represent 6.6% of global GDP and 8.8% of the world's population, yet receive only 1.12% of global AI investment. More than 90% of the region's high-performance computing capacity is concentrated in a single country. These figures expose a structural imbalance in who develops AI, controls infrastructure, and captures value. This is not lack of ambition. It is a gap in technological agency. The social consequences are equally serious. AI can expand access to education, health, public services, but it can also automate inequality when biased systems are deployed faster than public institutions can evaluate or correct them. The economic question is therefore not only how many jobs AI may transform. It is whether our countries will create new industries and capabilities or remain permanent purchasers of intelligence designed elsewhere. Culture and language are not peripheral concerns. Systems that fail to understand our language, institutions, and social realities will produce weaker outcomes regardless of their performance on global benchmarks. The AI for LAC regional dialogue stated the challenge clearly. Our region must avoid becoming merely a market, a source of data, or a testing ground for technology-shaped external priorities. We must move from adoption to agency. That means adopting models, developing solutions, governing strategic data, evaluating impacts, and participating in the standards that will define the next generation of economic and institutional power. This also requires stronger institutions. Ilya 2025 shows that 9 of 19 countries have national AI strategies, yet only a few have budgets, implementation plans, or impact indicators. The next measure of leadership should not be how many strategies we publish, but how many countries can convert policy into trusted, scalable, and measurable public value. Global AI governance. Thank you. Thank you, His Excellency. We'd like to call upon the distinguished delegate from Democratic Republic of Congo onto the stage, please. The distinguished delegate from Democratic Republic of Congo. Chair, ladies and gentlemen, it is an honor to take the floor before

you today in this prestigious room of the Palexpo. It's brought about significant change for the Democratic Republic of Congo and the Democratic African Nations, AAN. It is an immediate reality and a historic

opportunity is offered by it in order to transform our agriculture, modernize our health systems, and to dynamize our economies. AI history can't be written without Africa. The plan here in Geneva is very important. If there isn't any AI infrastructure, cobalt and cotton and copper, which are in the very heart of the DRC— Africa is not just an AI consumer. We are the very origins of its material. Therefore, our vision rests upon 2 pillars: opportunity and partnership, which is fair. We want to integrate this global supply chain of AI. This requires transfer of technology, which is obligatory, and the involvement of our youth and under national sovereignty. AI needs to bridge the digital divide. Not to broaden it. Also, the democratic application of ethics. For our country, AI should lead towards democracy and not a tool of oppression. We are addressing major risks: disinformation, manipulation of electoral processes, and the technological dependency. Therefore, we urge global governance of AI to protect national sovereignty and to protect our civil spaces. Ladies and gentlemen, the future of AI should be built with transparency and inclusion in mind. The DRC and the African nations stand ready to take on their responsibilities to ensure that AI is a true driver for freedom and for global justice. With that, I thank you. Thank you very much. To invite our final delegate of the day, the representative, His Excellency, of Ethiopia. Okay, in that case, um, Interpreters? Okay, thank you so much. Thank you very much for all of the interventions, and before we move to the closing remarks from our co-chairs, we'd

like to use this opportunity to thank the interpreters for staying with us for the extra 15 minutes, but you may now stand down. So with this, I'd like to hand over for the closing remarks to our co-chairs. First, I'd like Rashid, if you could go ahead, followed by His Excellency Mark Duma. Thank you. Thank you, Excellencies, colleagues. Thank you. Let me begin with gratitude to my co-chair, His Excellency Minister Dumba, for his partnership and steady hand this afternoon, to our moderators and panelists, to UNIDO, UNEP, ITU, and the Interagency Working Group who made

this session possible, and above all, to every delegation, every stakeholder who took the floor, and those who listened generously as others spoke. 3 hours ago, I asked this room for very— for specifics, and it got delivered. Here is what you told us. First, access is not the finish line. AI is within everyone's reach today, the fastest adopted technology in history, yet access alone does not create prosperity or jobs. Capacity does. The sharpest warning of the afternoon: the biggest risk is that we remain spectators while others capture the value. Second, the strategy is enablement, not a compute race. Build on digital public infrastructure. Make AI literacy a national priority, as Estonia has. And never underestimate distribution or transition. Getting AI to people and getting people through the change is where the real work lives. Third, sovereignty and openness are partners, not opposites. Local models in local languages on data that belongs to users, working with frontier models, not against them, because in AI, context matters more than anything else. Fourth, nobody wakes up wanting AI. People want better health, Better harvest, better public services. Success is measured in people, in lives improved, not models deployed. Let me end where I began. This afternoon, I offered Tim O'Reilly's test: create more value than you capture. This room sharpened it. The question is not whether AI can create more value than it captures. It's who gets to create this and who is left to watch. Spectatorship is not strategy; capacity is. This cluster's work does not end today at this time; it merely begins here. So thank you for giving this opportunity to me, and I'll hand it over to Mr. Dumba to take this forward. This has been an incredible day. An incredible first day in Geneva. Thank you, Rashid, my dear co-chair, for your partnership. It's been a pleasure. You have given us the what— 3 sharp priorities that will anchor our co-chair summary. Let me close with the why and with what happens next. I opened this afternoon by asking us to set new measures of success. This room

has answered better than I could have myself. AI will not be judged by how many models we deploy, but by how many jobs we create and how many jobs we uplift. Let me add, by how many languages thrive in the digital world and how many communities see their knowledge reflected, not erased. Those are metrics now. These are our metrics. Let us be held accountable to them. I also spoke of moving from big AI to small AI, of the opportunity of doing AI differently. The current path, which means ever larger models, ever more energy, and ever more water, is a race that even the wealthiest countries cannot sustain, and the rest of us should not want to run. But let me be clear, different is not a consolation prize. Different is the advantage. It means intelligence that speaks Fong, Mandarin, Creole, and Estonian. AI that runs on the infrastructure we have, not the infrastructure we're told to buy, and that serves the farmer, the nurse, the student, the entrepreneur. And the President of Estonia gave us a compass today. Do not race to be first on AI, race to use it wisely. That is a race every nation in this room can win. In my opening, I also said that we are limited only by our imagination. We have applied plenty of imagination to invention. The test of our generation is to apply to distribution, which means to who benefits, where, and how fast. And it is a generational test. There has never been a moment with more talent, more technology, more capacity to act, nowhere more so than on my continent where the median age is 19. Let me thank our panelists and moderators and everyone who took the floor today. our reporters and the Secretariat, and you, Rashid, once again, my dear co-chair. Now, a challenge to everyone in this room, every delegation, every company, every institution. Please do not wait for our summary. Please, before you leave Geneva, take one of those 3 priorities Rashid has set out for us and give it a name, a date, and a budget. One policy, One partnership, one palette. I began today by speaking of networks, net worth that exceeds the wealth of nations. Look around, we have one, and the network right now is judged by what it delivers. Tomorrow, Rachid and I will carry your voices into the dialogue of dialogues, and when this dialogue meets again in New York in 2027, let us not report on what AI could do, Let us stand up, one by one, and say what we did. The work starts now. Thank you. Very briefly, just thank you for your valuable contributions today, and of course a special thank you again to the thematic cluster co-chairs for capturing the key outcomes and reporting back on them to the main plenary. in tomorrow's Dialogue of Dialogues. We look forward to reconvening

to turn these insights into action. I wish you all a nice evening. Thank you very much. A round of applause.