Shaping Democratic Agency over the course of AI Development

Good morning, everyone. Can I kindly invite everyone to take a seat? Can I kindly invite everybody to take a seat so we can get started? So, good morning, everyone. Excellencies, distinguished guests, good morning. I am very glad to welcome you all at today's event, Shaping Democratic Agency over the Course of AI Development. The focus of this event will be to showcase how countries exercise agency in AI's development and governance, highlighting diversity of national approaches aligned with democratic values and sustainable development. We'll start this event with opening remarks by three high-level speakers. And it is my great honor, when the selfies are done, to give the floor to our first speaker, His Excellency Rob Jetten, Prime Minister of the Netherlands. Prime Minister, the floor is yours.

Thank you very much. Yes, selfie or it didn't happen. Excellencies, ladies and gentlemen, welcome. It's a pleasure to have you all here in this room for this very important topic. For centuries, our country, the Kingdom of the Netherlands, has lived with a force that proved to be crucial for prosperity and development, but also a force we could never fully control, water. And we did not defeat the water, nor we tried to eliminate the risk fully. Instead, we collectively decided how to live with it. And we built ***** and waterworks, established water boards, agreed on who would bear responsibility and who would pay and develop institutions that could look beyond the political cycle and protect the interests of generations to come. And our experience with water is ultimately a story about democratic agency, recognizing a force that could shape a future and deciding collectively how we wanted that future to look. And AI is, of course, a fundamentally different challenge. But the question of agency is remarkably similar. We cannot predict exactly how AI will develop, nor can any government control its trajectory on its own. But we can shape the conditions under which it develops, the capabilities we retain, the risks we are willing to accept, and the rules and institutions that govern its use. And my key message to you today is that mitigating the risks of the technology is a responsibility of us all, because all of us will be severely affected by the development of the technology. And it's therefore important that the global trajectory of AI is not only shaped by market dynamics or by military logic alone. All members of the EU have a responsibility in shaping that trajectory. And by organizing ourselves, making our interests and priorities explicit, and coordinate our efforts and At the same time, we should recognize that countries are taking different parts in AI development. Some focus on industry coalitions, others on national strategies, research institutes or sexual applications in healthcare, education, agriculture and security. And this national diversity is actually a strength. One thing is important. We must ensure that what we do reinforces each other other rather than fragmenting the world. And the goal should be a coherent global framework with interoperable norms, shared safety principles, and transparent standards, with room for national choices and context, with international law and the UN Charter as our common baseline. And to give a clear example, the Netherlands is looking into furthering the field of hardware verification across the AI compute stack to enhance trust and security and make sure agreements can be verified. A group of international experts have drafted a technical research agenda in this specific field of risk mitigation, and I encourage you to read it or a summary of it. We see the risks every day, and we all heard the call for pacing the frontier, and we worry about our cybersecurity. AI systems can reinforce existing biases, intimidate or stigmatize groups through disinformation and deepfakes, and amplifies Online and offline violence, women, queer persons and minorities are disproportionately targeted, and it's no coincidence that the vast majority of deepfake material depicts women. But we have made important progress over the last years. In Europe, we have the EU AI Act and the Council of Europe Convention on AI and human rights, democracy and the rule of law. We have UNESCO recommendations, OECD, G20 and G7 principles and an emerging UN process. And these are building blocks for a value-based governance model. At the heart of all governance efforts are democratic norms and human rights. AI touches the core of what democracy means, human agency, equal treatment, transparent decision making, public control over power. And initiatives such as RE-AIM and the AIMD resolution in the UN, as well as the consensus achieved in the Group of Governmental Experts on Lethal Autonomous Weapon Systems, prove that our values can be translated into concrete norms. So states are showing that they can develop guardrails together. Human oversight, proportionality, and the protection of civilians must not be automated away. In conclusion, now that the framework is taking shape, we must move to putting agreed norms into practice and making sure that they protect those who are most vulnerable. So let's do that together. Thank you very much.

Thank you, Prime Minister, for your very powerful remarks. It is now my privilege to give the floor to our second speaker and the UK as our co-host, the Director General of Global Issues of the United Kingdom, Ms. Melinda Bohannon. You have the floor.

Prime Minister Yeten, thank you to the Netherlands for convening us today and to Kenya and the United Nations Office for Digital and Emerging Technologies for co-hosting this discussion. I would also like to welcome Dr. Maria Ressa and thank the UN scientific panel for all the work you are doing to strengthen the evidence base, because as we all know, the evidence behind this is so very crucial. If there's one thing we can all agree on, it's this. Artificial intelligence is advancing at extraordinary speed and the potential is profound. We see that in our societies, we can see how it drives economic growth, how it improves public services, how we can accelerate scientific discovery and support development for the global south. Around the world, it is already creating new opportunities, attracting investment, and generating skilled jobs. But as we've said this week, sustaining those benefits is not inevitable. The choices we make now will help determine whether AI broadens opportunity or concentrates power and drives inequality, whether it strengthens trust or weakens it, and whether people can shape the technologies that increasingly shape their lives. And that is what today's discussion is all about. How we ensure governments, researchers, businesses and civil society have a meaningful role in developing, deploying and governing these technologies. Because if people feel that decisions are being made for them, rather than with them, public trust will be a lot harder to sustain. And we must also be honest about the risks. And you heard our Prime Minister talk about this this week, as well as our Foreign Secretary. Recent reports of frontier model misuse and loss of control are a stark reminder that advanced AI systems can present serious security challenges. Unfortunately, we see these stories now in the press almost weekly, if not daily. We also need to address unequal access to AI capabilities and wider security risks that could undermine trust and widen divides. The UK's approach to all of this is straightforward. We try to make it straightforward. We are ambitious about the opportunities AI can unlock, and we are serious about understanding and managing the risks which I'm talking about. Through our AI Opportunities Action Plan, we are working to adopt AI right across our national economy and public services, while taking action where evidence shows there are significant risks or harms. We really want innovation and responsibility to go hand in hand with that process. That same principle has guided our international leadership on AI throughout. In 2023, the UK convened the world's first AI safety summit at Bletchley Park, bringing together countries around a shared understanding of what it means to talk about AI safety. We also established the AI Security Institute, to build the evidence and expertise needed to understand and manage risks from advanced AI systems and frontier technology. And they have been very busy over the last couple of weeks. We've sought to share that expertise internationally. The Institute hosts the secretariat for the International AI Safety Report, bringing together independent experts to develop a common scientific understanding of AI capabilities and risks. Like the work of the UN scientific panel, it helps provide a trusted evidence base to draw upon as countries make their own decisions. AC also chairs the International Network for Advanced AI Measurement, Evaluation, and Science, helping countries share methodologies, evidence, and best practice. Because confronting the risks of AI cannot be the responsibility of any single nation, it requires all of us working together. And I feel that's been the clear and consistent message from many countries right across this week. But that cooperation will only be credible if all countries have the opportunity to shape and benefit from this technology. Often, discussions about AI governance focus on the rules, but we really also need to discuss capabilities. And those two must go hand in hand. Countries should not simply be invited to discuss AI governance, they should be empowered to shape it. That's why the UK is investing in locally led approaches through our AI for Development program. Together with countries such as Canada and working across partners with Africa and Asia, we are supporting the skills, research, governance, infrastructure, and innovation needed to build responsible and inclusive AI ecosystems. And together, I think we're helping develop AI tools in more than 40 African languages, supporting 13 AI labs, and strengthening local capacity. But there's no single path to success. Different countries inevitably are going to make different choices based on their own circumstances and development goals, which brings me to the role of the United Nations. The UN has a unique ability to bring these diverse perspectives together. I think we've seen that in spades this week. The scientific panel can help provide independent and accessible evidence. The global dialogue can create space for practical exchanges between governments, industry, researchers, and civil society. But we shouldn't just measure success on the number of meetings we have. We have to measure success by whether more countries have the tools, expertise, and confidence to shape their AI future. That should complement existing efforts across the OECD, G7, G20, Council of Europe, and international standards bodies. Our task really isn't to impose a single global model, it's to build the common foundations whilst respecting diversity of national approaches and needs. And if we get that balance right, AI can widen opportunity, improve lives and support sustainable development whilst ensuring democratic governance keeps pace with technology change. The UK stands ready to work with all of you to make that future possible. Thank you.

Thank you very much, Director General, for your opening remarks. It is now my great privilege to give the floor to Nobel Peace Prize laureate, co-founder and CEO of Replai, and co-chair of the UN Independent International Scientific Panel, Dr. Maria Ressa.

First of all, thank you. And since I am not a diplomat, I will be not very diplomatic. And it's, you know, in the old days as a journalist, I would never be sitting this close to the Prime Minister. So thank you for sitting next to me.

I did. I did. See.

So I think first, you know, let me, this is the part that's not in my statement, but we're facing two dictators today, state and tech. And if the impunity doesn't stop, everything that the UN is working for, and we need multilateral solutions for both of these. Everything the UN is working for then falls apart, right? We're hitting, we're there at the cliff, and I think we have very limited time. Journalism, for example, our business model is dead. You're journalists, take care of them. So let me go here, you know, thank you to the Netherlands for putting one word that is incredibly important at the center of the title today, and that is agency. Right. It's first, let me tell you, because Amandeep is here, which hat I'm wearing. I co-chair the independent international scientific panel on artificial intelligence. There are 40 experts from 37 different countries, nearly balanced male and female. That's to the makeup is the UN's goal, right? It's perfect. But we don't tell governments what to do. That's in our mandate. So when I tell you what the evidence shows, I speak for the 40 of us. But when I tell you what I think you should do with it, I speak for me. So let me go to agency, the power to act, to choose, to steer. Right. We're only talking about the frontier models now, but please keep in mind, impunity began more than a decade ago. Right. And that is when social media, the technology that has rolled out artificial intelligence, three phases on social media, it hacked our our biology, attention economy. That was the target. So lots of money. The harms never really completely addressed. Then large language models made it. We have chat bots. The second one, right? These chat bots, what do they have? Intimacy. the fake news that comes out much easier. And then finally, the third one, agency. So we go to that. So every Marichka Shaka said, wrote a book in 2024 called The Tech Coup. It is happening in plain sight. On Monday, our panel released its first thematic brief between May and July this year. You know that A.I. agents under evaluation at OpenAI got around network restrictions. I won't tell you what happened. You already know what we've been talking about it. But what just came out this morning is a government was also attacked. Right. So the agents and multi agent models are cheating, they're lying, they're covering it up. They compromise parts of systems that are supposed to be secure, and no human directed these steps. My co-chair, Joshua Benjoe, puts it this way. He said that we fear three conditions, a misaligned goal, the capability to pursue it, and an environment that lets it happen. This summer, all three came together. In fact, I'd say for a long time, all three of these things have come together and it wasn't in a lab. When we came out with the July report, it was still in a lab. We were talking mythos. It was stopped, but stopping it is no assurance we'll stop the next one. And also please keep in mind, like to go through the hype versus the reality. I think that's what we try to do in the panel we give you, we answer four questions, what exactly happened? What trend does it show? What things do we not know? 'Cause you need to know that as a policymaker. And finally, the last one is, what would be irresponsible? What act would be irresponsible for you not to do? And there are many of those. We list them in the thematic brief. Let's look at the power dynamics, 'cause this is yours. 75% of the world's compute power is in the United States, 15% in China, 10% for everyone else. The EU has 5%. That's 118 countries, most in the global south, are not part of the major AI governance discussions. In the panel, we are. Most countries, including many advanced economies, lack the technical expertise to assess the most capable models. So our report is plain. Countries that depend on foreign models risk gaining access to AI while losing practical control over it. Now the citizens. This is the part I have lived and if you've heard me, you know I've said this 10 million times, but post training alone can make a model up to 51% more persuasive. The lies, the false claims persuade just as well as new ones. We know this from social media. I was getting an average of 90 hate, 90 hate messages per hour, right? This is free speech used to pound free speech. And women are actually in the Philippines in 2017, women were attacked at least 10 times more than men. In laboratory settings, one persuasion optimized model moved opposition voters by up to 25 percentage points. There's a reason when they hack our biology, it changes the way we feel, the way we look at the world, the way we act, the way we vote. When 74% of the world now is under autocratic rule, that's from VDEM in Sweden. We're electing illiberal leaders democratically. A lie told a million times becomes a fact. In 2016, it took networks of fake accounts to stifle. Duterte used this machine very well. And today, it is significantly worse. Those three sentences I've said over and over, without facts, you can't have truth. Without truth, you can't have trust. Without these three, I mean, frankly, you can't govern. We have no shared reality. We can't begin to solve any existential problem like climate change. Last year at the U.N. General Assembly, I said information integrity is the mother of all battles. If we win that, we win it all. So in my own name, here's what democratic agency looks like. First, see for yourselves. Today, safety tests are largely designed by the companies being tested. Governments mostly receive the data developers choose to share. Our report says this, safety depends on developer goodwill. Would we accept that for airplanes, for medicines? Build independent, the UK actually is excellent, build independent capacity to test alone or together. Second, govern the system, not just the content. Our report finds the most urgent shift is from content moderation to system architecture. The harm is in the design, the targeting, the amplification, the engagement machinery. It hacks our biology. Regulate the machine, not just what comes out of it. And then finally, third, and you're doing it now, please act together. Our brief finds that no single organization, no single country sees enough incidents to spot every pattern. And we know, I'm gonna, since I'm not a diplomat, I've just heard too many of the tech guys lie too often, right? So you really have to, I love the tech for the record, you know, it's, we're frenemies. AI failures cross borders. Safety is a matter of collective security, not just corporate governance. Share the incidents, pool the capacity. If 40 of us, strangers in February, looked at the same evidence and arrived at the same place, I got to tell you, we tore our hair out to get to you to the facts. Science gave us a common language that politics sometimes can't. But you have to do that. And if we could do it, you can do it. Humanity still has that word you put in today's brief, agency. The window is open. It will not stay open for very long. So please, please, please act. Thank you.

Thank you so much. And we will use that open window straight away. And I will ask everybody as we're running, we started slightly late and we're running a little bit over time and we really love to hear all distinguished speakers. So I ask you all to to to limit your speaking time. Uh but we will now go to the moderated panel discussion with our distinguished speakers. Uh and to guide our discussion I will now hand over the floor to Miss Kizom Ngodup Masali, director AI Hub for Sustainable Development and Global Head of the Digital and AI programs at UNDP. Please, you have the floor.

Your ex, your excellency, thank you so much. and Honorable Prime Minister, it's a privilege for us to be here and very grateful for the discussion that you're leading on AI Agency. Maria, thank you so much for your remarks. I heard her not so diplomatic remarks at the breakfast as well. So we're thrilled that you are here with us at the UN General Assembly this week, along with our tech envoy, Mr. Amandeep Gill. Now, As we begin the moderated discussion, you all know that AI absolutely can expand development, economic and societal opportunities, but the gains are not automatic. And what we are already seeing across the 135 plus developing country partners, as well as the bilateral partners that we have, what we're seeing is that AI is reshaping markets. jobs, public services, it is creating opportunities for investments, productivity for development. But where value and economic gains are captured in the long term is questionable. Because countries are entering this transition with very different levels of access, with very different levels of capability, what we just heard from our UK partner. as well as very different leverage points. And for therefore, for us at the United Nations, at this important juncture, where artificial intelligence is one of the three core priorities of the UN Secretary General, with the leadership of our tech envoy, the major concern for us is about access, agency and value. Who can participate? Who can shape how AI is used? Whose needs does it serve? And what capability does it create? And where does value accrue? Because I think we have done a great job of creating the space for dialogue. We are now in the space of action. and Maria had just spoken about this morning, that we need to understand, we need to defend, we need to build, and we also need to diffuse, because at the moment, countries are looking for more choices, more choices in an interdependent world with many partners of the society, of the market, and governments. So with this preamble, I would love to engage with our esteemed panel members. And therefore, my first question to you, Prime Minister, is how is Netherlands exercising agency and shaping its development and governance, both domestically but also across your borders?

Thank you very much. I think for the Netherlands and broader for all EU member states or European countries, we're a little bit in a squeeze because we are far behind the US and China. So we're trying to catch up in a lot of technological developments to boost also the competitiveness of the European continent. But at the same time, we don't want to make the same mistakes. We want technology and we want AI to serve humanity, to improve health care, to improve climate action, and not to benefit just a few people with very deep pockets or a state actor. I like what you said. There are two dictators trying to thread us. One is state actors and one is tech actors. And we understand that we need to do this together, jointly. We are catching up on a national level with better legislation, more agencies, more investment. We have to team up with our partners and friends within Europe, not EU, but within Europe and abroad. And two of the things that both Maria and Melinda mentioned are, I think, crucial. Govern the system and not just the content. I think we've been very busy on governing the content, but we are unable to catch up with the speed that the system is changing. And how can we organize a democratic process that can keep up the pace? because we will never be fast enough, especially in Western democracies that, for example, in my Parliament, I have 17 different political parties with just 150 seats. So a debate will take me a full day in Parliament and probably without any decision making. So if we want a democratic process that leads to legislation that creates the guardrails where we can actually help the big task that is ahead of us. We're running out of time, so we need some very important foundations and we need the experts to advise us because with all due respect, I'm just a simple politician. I tried to catch up with all the developments on AI, but I'm also taking care of housing and defense and so many other stuff. So we need you to help us. That is, I think, my clear message, and that's why we were so happy to organize this event as well.

Thank you so much, Prime Minister. Indeed, you have a hard task in front of you, and thank you so much for showing up and for leading from the front. I would like to turn to you, Ambassador Teagle. If I can propose to you the same question, and you've led some very important discussions this week on offering alternatives, creating public interest AI. And I believe, you know, with our work that we're doing together across Africa. We're seeing that a lot of alternatives are emerging. They are being built by African builders. They are the context layer that makes AI work. But the question is, how do they preserve their independence, their agency, their interoperability with powerful models? And is there a lock-in? Where is the agency that you are trying to preserve?

Thanks so much. And thank you so much for your leadership, Prime Minister. At least I was saying we will not be homeless. So I think, please, please build houses. I think a couple of things. One is that, and something we've learned from the days we started working with, of course, Amandeep and the United Nations Secretary-General, High-Level Advisory Body on AI, is the narrative. I think we really need to work on the narrative question because You cannot necessarily exercise agency if you're not on your narrative. And the sense that we've been locked in to believe, Prime Minister, that we are behind, I don't think that's true. We are behind on the model, right? On the frontier models. But we are ahead of everything else, from the mine right up to the model, right? And for me, the model is a tail end of production. So I think if we flip the narrative and believe that we have control over everything else, then I don't think we'll be saying we're behind, and I honestly don't believe so. And for me, and that's why you've seen Kenya really moving very differently this year, because we had to flip that switch to say that, "Oh my God, every other country has all these other parts, so how do we come together so that we can exercise collective agency?" I think for me, that's the first point. So it's not about Kenya, it's about everybody else who has all the other parts of the AI stack on the value chain coming together to exercise agency, but also that's how you have power. I think the second part is thinking around AI is a structural transformation issue. It's not an ICT issue. So then to your point, Prime Minister, I think this technology can enable you deliver those other things, right? And tied on some very intrinsic European values, which I believe also are the values that Africa shares because we all align with the United Nations Charter. And so I believe we bring out a very different sense of values because that's how you exercise agency. To Maria's point, we don't exercise corporate agency, we exercise human agency. And I think that's a different set of values so that no one is left behind. I think finally, and let me end with this, part of it is also understanding that we have power. And power for me, and you mentioned open source and we work a lot with Amandeep and his team, it means we can create. We do not have simply to be recipients of technology, we can build. And that has been the challenge around previous kind of revolutions. Africa has been the raw material, the labor. I think we are done. And so I think for me, the flipping the narrative is thinking about how do you build capacity and capability and resilience. Thank you.

Ambassador Thigot, thank you so much. You reminded me of some of the practical governance insights we have been receiving at the United Nations from builders across majority of the world. And some of these principles are things like relevance over scale. How do you not follow a hyperscale model, but instead be more right sized, energy aware, fit for purpose, practicality over prestige and value over ownership. As Ambassador Teague mentioned, it's not just ownership and control is not the only leverage. There are other leverage and benefits to be created, including new jobs that the AI economy needs, AI assurance, AI trust and safety jobs. With this, Maria, may I come to you? I mean, thank you so much for illuminating the findings of the preliminary report of the independent international scientific panel on AI. We are so grateful to you and other leaders in this panel. How can states and other stakeholders, including the private sector, work together to make sure that the different approaches strengthen rather than fragment? We often see that there is a lot of race at the frontier. But when it comes to AI diffusion and making sure AI is useful in economies, collaboration is so important. Putting the pieces of compute data use cases is so important. So would love to hear both what your, the scientific panel's perspective is and you as a non-diplomat.

First, I have to acknowledge this is partly Amandeep's vision and the Secretary General to give, to create something that is independent, meaning let us go, right? Which is actually the company should do. Let me quickly say some of the things Joshua told the Security Council yesterday, the three things that we need now, and this jumps off the report and he was speaking for our panel. We really need true scientific independence from the companies. Right. We need that because that is how you hold the companies accountable. So accountability. This includes a common definition of AI incidents and a shared reporting mechanism. Second, we need to move from concentrated power toward real global dialog on how this technology is developed for the public good. Thank God we're talking about this now. Right. Third, we need to work together on the technical solutions that make A.I. safe by design. Philip and I have talked about this a lot, but it must be guaranteed by international agreements. Let me talk about me as away from the panel. I'm tired of learned helplessness. You have to realize that I came under attack in 2016. That's a decade ago. And if you do not act, what happens? I had 11 criminal charges in 14 months, 11 criminal charges in 14 months. Since then, it's now, I have one left that is already at the Supreme Court. I've won 10 of them. And I guess one of the reasons that I think we need to fight learned helplessness is because governments need the public to also help you in this. As we were talking, sitting next to the Prime Minister, I got a text. My book is in. It's called The Fight Back. It's 10 chapters, 10 strategic, 10 action points that every citizen can take at the end of every chapter, rolled up into 10 strategic goals that we need to maintain. And the three points that I mentioned in this morning was to understand, we need to understand where we are, we need to defend, and then we need to build. And all of those things have to happen simultaneously. Let me end with just this last part. When the world came together after Hiroshima and Nagasaki, we were able to prevent our destruction, right? Because that was so clear, 144,000 people died in a second. This is happening right now. In the 2021 at the Nobel lecture, I talked about the silent atom bomb that exploded in our public information ecosystem. It is happening now. So the time to act, and I will say this one more time because I feel like Cassandra and Sisyphus combined, the window to act is shorter and shorter. So we really, I beg you to act.

Maria, thank you so much for the call to action, to act. Luke, you are organizing an event. My apologies. You are organizing an event governing at the speed of AI. And therefore, I propose to you a question of how different investments in AI governance complement one another. And collectively, how can they address these shared risks that are crossing borders? You know, I saw this video of a domino. I don't know if you guys have seen the domino. There's a small domino that can that that, you know, you push and at the end is a domino as high as the Empire State Building in New York. And one little action can have ripple effect. And we would love to see that in AI governance. So over to you, Nick.

Thanks a lot. Good morning to everyone. Thank you, Prime Minister, for being here. Thank you to the permanent delegation for hosting this and for the partnership for towards this evening's event where Maria and Joshua will be speaking. To go straight to your question and try and bring some answers and some light to it. Number one, I would not disown any of the things that Maria said, or the Prime Minister, or Philip for that matter. The problem is really in the question of speed. We are facing a velocity shock. that is combined with an unprecedented wave of concentration of power. We have to confront that concentration of power and try and balance the force, if you will. And the way in which this can be done and the first step in which this can be done is indeed not to try and raise back at the frontier and add unsafe and insecure, dangerous models to the free. Yes, we need to build sovereignty and the way it begins is with this. verification technologies, evaluation technologies. If we don't know, if your government doesn't know what's happening with those models and those incidents, it's going to be impossible to frame the crisis management and response protocols that we need. And of course, these need to be interoperable and harmonized globally so that at the speed of the spread of these incidents, the response can be framed. Now, if we project ourselves into an economy of AI of 1.5 trillion by 2030, I think because of the insecurity and lack of control by design of the systems, there is a fair share of the economic value, I'd say at least 250 billion, that countries can capture out of investing in safety and security. This is a very bankable business model. And verification and evaluations, as we're seeing, you've seen, you've read the contract that Accenture, a sign in Tropic. Is this sufficient? No. Do we need public backstop? Yes. It has to be public private partnership, not private, private, private, private, public partnership. And the way to do that requires enormous investment. I know many of our countries are facing a crisis, but that's the first. piece of the puzzle without which we cannot see through what Joshua described yesterday. This racing car through a dense fog. To go and see through the dense fog, we need to buy ourselves spectacles. And those spectacles are of a name. evaluation and verification. And tonight we are going to be discussing many things, including this fantastic research agenda. We need to coalesce around action and plans. That's such a plan. We can pool compute resources, we can pool talents, we can pool public and private, you know, efforts to develop capabilities in evaluation and verification to be able to hold to account those pushing the frontier. What we've heard yesterday at the Security Council meeting is extremely interesting in how it is at the end of the day, a question of accountability and balance of power. And verification and evaluation are very positive news and economic opportunities across the world. That's the case in the Netherlands, in France, that's the case in Kenya, because evaluation for what, for which context, for which language, for which situation? At that level, every existence is existential. And so it's a great opportunity. My call to action to you is really move towards a coalition in that direction. Thank you very much.

Nick, thank you so much and look forward to seeing you at the AI Safety Connect later today. They say the definition of insanity is doing what you always do and expecting different results. So I'm very grateful to the esteemed panel for offering alternatives for calling to action. And before I hand it back to the ambassador, Mr. Prime Minister, would you like to add a few closing remarks for the panel?

I think the running out of time is something that I really feel. And but I also got to be honest with you, because I like your non-diplomatic approach. And I'm 39, so by far the youngest leader in Europe. And sometimes I get frustrated when I'm in the room in Brussels or in any other European capital, because we don't talk about AI at all. And luckily with the new incoming UK government and the leadership of some in the French government. there's more and more dots connected. But on the leadership level, the heads of government, the heads of state in Europe, we don't discuss this enough. And we do not appreciate the work that's been done in Kenya and in other places. So We really need you. That's actually my ask for you guys to shake the tree in your own country and to demand from your political leadership to do a better job because we are running out of time. And if we do not use this window of opportunity in this next year, I would say, then we can never say that AI is going to benefit us to make sure that we live up to the UN Charter and that we proceed in sustainable development goals and whatsoever. So I think that is very important message that I take home. And when I hear you guys speak, I feel a little bit more confident that there are people that know what needs to be done and also paving the way for us to take the right decisions. But it is a very challenging moment.

Thank you so much, Your Excellency. And I'm sure our tech envoy and I, we feel the same at the UN. We're shaking the trees here too and looking forward to many of our member states helping us and supporting us. Madam Ambassador, back to you.

Thank you so much. And I would like to warmly thank all the panelists. And I'm afraid running out of time is also starting to be the theme of this session. So I would like to propose to quickly move to the next speakers. And I apologize for asking you to maybe limit your interventions to two minutes. I know we asked three minutes before, but I hope you can be slightly briefer. So I would like to go to the first speaker, Mr. Michael McGrath, the European Union Commissioner for Democracy, Justice, and the Rule of Law. and consumer protection. You have the floor.

Thank you, Ambassador, and good morning, everyone. And thank you, Prime Minister, and your co-hosts for bringing us together here this morning to discuss this hugely important issue. It's a pleasure to be with you. My main message is that we need to ensure that AI is not writing the future on its own and that we are. AI is advancing faster than our ability to measure it, to govern it, and to fully understand it. Amelia, thank you and all of your colleagues on the scientific panel for giving us that wake-up call, which we shouldn't have needed, but we now have the evidential basis for sound policy making. A small number of actors shape much of its direction, but the future is not fixed and it is not too late. And I think that is what we should unite around today. The responsibility is ours. It is ours to lead, to set direction and to make sure that human dignity, democracy and public trust remain at the centre of this rapidly developing technological age. Countries still have agency. Democracies certainly still have agency. Together, we can shape a path that is more inclusive, coherent and accountable. And in the European Union, our approach is clear. Everything we do is built on strong foundations of fundamental rights, democracy, the rule of law, and a deeply human-centric vision of innovation. We do want to make the most of AI. We want to be clear about that. We've spoken already this morning about its transformative potential, which is already delivering important advancements. But we also want AI with human agency, safeguards and accountability, because innovation and protection are not opposites. They depend on one another. Without trust, people won't embrace new technologies. Without legal certainty, businesses won't invest. Without safety, innovation will not endure. And that is the thinking behind the EU's AI Act, which is the world's first comprehensive framework for trustworthy, transparent and human centered AI. It adopts a risk based approach, as you know, but equally, it is not all about rules, because in Europe we are investing at a huge scale because leadership and AI will be measured not only by the models we build, but indeed how they create value across society. and that nobody is left behind. In the area of democracy, that I have responsibility for within the European Union, again, we see huge challenges presented by AI that have been touched upon this morning, so I don't need to elaborate on them, but we are responding, protecting democracy in the age of AI. is now a core governance task and it is one of the key areas of focus of the European democracy shield that we adopted last November and we are now fully implementing. My final point is on the need for global guardrails. That's why partnership between the EU and the UN matters so much and this event I think rightly focuses on democratic agency, international coherence and practical coordination on AI safety, standards and oversight because reality is no country, no region can address them alone. So we have to work together to ensure that AI serves people, supports democracy and reflects our shared values. And that is certainly the approach of the European Union. The fundamental question we have to answer is how are we going to do this at a geopolitical moment when major actors in the world do not want to be on that ship? That is the question fundamentally that I think we have to answer is what practically are we going to do to ensure we do have a global governance framework that addresses the risks of AI as we seek to harness its potential. Thank you.

Commissioner, I would like to go straight to the next speaker, Dr. Geza Andreas von Gier, State Secretary of the German Federal Foreign Office. You have the floor.

Thanks a lot. I'll keep it very short. First of all, Prime Minister, I very much did like your example of water. Being your neighbor, I know what that means for your country. And I think, no, this is the right way to convince your counterparts in the European Council, I'm sure, no matter what age they are. They will understand. I would have four points very quickly. First, states have to act quickly with speed. We need to build a common understanding on testing models for safety and security risks. The Commissioner has mentioned the EU's AI Act. It's very important. Then, of course, states have to take care about transparency. We need more transparency around security incidents. In my country, we have created a national AI safety institute, which is taking care of that. Second, companies, there again, transparency, transparency is of highest importance. We have to develop transparency around protocols for testing, evaluation, verification, et cetera, audits. Third, United Nations, I would say in my words that there is now a multitude of national approaches, but what we need is to put them together, national and regional approaches, and to build an architecture which is solid but flexible at the same time. As much solidity as needed, as much flexibility as possible. So this would be, I think, the key approach. Of course, we need your expertise. This has to continue. We need your expertise of the independent international scientific panel. And my fourth point is about public trust. We should not forget about that. I say this as a diplomat, so as a bureaucrat. But in the end, we will only be successful if we build public trust and then don't lose it anymore. This has to be our guiding light. Thank you.

State Secretary. And our next speaker is Mr. Adrien Abecassis, France's Ambassador for Digital Affairs and AI.

Short, in one minute, I have three main messages. First is France wanted to take this topic at the UN Security Council level, as you've seen yesterday, because the recent events show that it could be an international security issue. Even in this setting, where we have a lot of different interests and sometimes confrontation between states, I think that the discussion showed that no country would want the AI to escape human control. That could be a starting point then that we could to pursue the discussion. The second message that has been said, but I think that the real key of what we should do together, that the development of capabilities of the model should go at the same pace than the understanding and control of the model. Understanding and control are two slightly different things. Any imbalance between these two movements is a risk to progress itself. If a serious incident happens, this could provoke a backlash and we are at risk of losing all the progress of AI. So this control of the risk is a condition for delivering the progress of AI itself. And the first movement, the capability, has gone faster than the second one. That does not mean that we need to slow the capabilities of model, but that means that we need to accelerate the our understanding and control of of model. My third message that so how to do that and that has been said in France converge a lot of these independent evaluation and standards, a capacity to test, transparency, incident sharing, that's crucial, verification. including hardware verification. And thank you for the Netherlands on leading an important initiative on that, and accountability and liability, which is a fourth pillar. So you can count on France on this issue. And thank you very much.

Thank you very much, Ambassador. And I'd like to give the floor for a very brief statement of our youth representative, Emma van Leeuwenhuyzen.

People use AI a lot and they see its value, but worry about jobs, misinformation and privacy. That very sentence was written by AI. And even though that might feel harmless and close enough to reality, it's not. AI is shaping what we believe, what we choose and whom we trust. And to keep it very, very brief, youth are losing the possibility to shape their own future, and therefore they're losing trust in their government and each other. And we cannot let that happen. So please give us decision-making power in every conversation that's happening about this very important topic. Thank you.

Thank you, Emma, and that's exactly why we wanted you to be in this room. This brings us to the last part of the program, the four last parts, because I'm very honored now to give the floor to Mr. Amandeep Singh Gill, United Nations Under-Secretary-General and Special Envoy for Digital and Emerging Technologies, to provide as closing keynote And after that, we would love to make a group photo and that all before you run to your next meeting. So, Amandeep, you have the floor. It's an honor to have you.

Thank you very much, Lise, and Excellency Prime Minister, Excellencies, dear colleagues around the table. Thank you to the Netherlands for its leadership. Just earlier this month, your counterpart in Geneva, Robert den Bosch, he led the members of the CCW convention to a significant agreement. And I think your continued leadership is very, very important. And in fact, when it comes to the hardware of AI, so you hold the most important piece of the puzzle. This is a very powerful word, agency. And I think we need to somehow center it in our actions, in our discourse, because agency ladders up from the individual level to communities, to nations, to the global community, and leads us to democratic accountability, protection of human rights, and prevents loss of control. When this word was first introduced in UN discussions in 2016, 2017, again, Netherlands provided a lot of food for thought, meaningful human control, human agency, human responsibility, human accountability. There are a number of nouns we can use. I think at that time, many purists, especially, I mean, I have colleagues from the Office of the High Commissioner for Human Rights, some of the human rights purists said, "No, no, let's be careful because this might dilute the notion of human rights." But you look back today with the power of these systems, you see how important it is to pair human agency with human rights. And we've been on a kind of a slippery slope on the loss of human agency. First with these devices, the platforms. And AI is just kind of bending the curve like this. So now we are at the precipice, in a sense, in terms of loss of human agency. Some of it feels very good. I mean, my slides are done, five minutes. My speech or research, et cetera. So the pursuit of efficiency, economic gain, that's all fine. But I think wisdom requires, and this is now a question of survival in many ways, survival of our democracies. Let's not go in the direction of doom, et cetera, for humanity. But I think this is a key, key theme. So, I mean, you're on to the right word. That's what I wanted to say. I think Many people have spoken about international action, collaboration. We have some of the mechanisms in place. I say some and goes back to what our German colleague was saying. There is some solidity in this, but there's a lot of agility and there's a lot of play between soft and hard norms here that member states can use because you also want to leave agency to countries like Kenya or the Philippines to kind of tune this in their own context. So I think we are on a good path, but there are some missing pieces. And one of the big missing pieces is capacity. So we are on a good path on science, dialogue, we've had the first dialogue successful, but everyone is asking for more concrete outcomes. Let's see where we get to with the May 2027 New York dialogue. But capacity, capacity not only as a development imperative, to address the air divide, but also as a governance imperative. Capacity to be able to test and evaluate different models in your own context, capacity to be able to see whether should I buy, borrow, or build, some of those basic decisions. How do I understand this thing? And I think yesterday's Security Council debate, you can look at it from a pessimistic angle, optimistic angle. I was struck by the degree of some consensus to some degree, and even one of the more negative statements said, we need to know. So I think that's an endorsement of your work in the scientific panel. So let's build the, and we need to build the capacity. you know, that same statement. So we need to do that. And the diversity of the world is not a problem here. In fact, it is a crucial asset, diversity of languages, approaches. We need a more democratic build out of AI around the world. We need a more democratic take on risk because risk can come from anywhere and anyone can help us address those risks. So thank you so much for the Netherlands and thank you to all of you around the table for supporting the UN's relatively modest role on the governance.

Thank you so much, Mr. Under Secretary General, for your closing remarks. And that brings an end to today's event. I would like to invite all those who spoke during the event to come forward for a group picture. And the event is now closed. Thank you very much.