Multi-stakeholder plenary meeting with a high-level governmental segment: Dialogue of Dialogues - Reflections, Complementarities and the Way Forward & Closing Ceremony

Tómasdóttir argues that while many governments, institutions, companies, and civil society organisations have put forward important AI principles pointing towards safety, inclusion, transparency, and human rights, these principles have not yet translated into shared guardrails or common standards. She warns that stating responsible-sounding principles while avoiding the difficult choices that responsibility requires creates a false sense of security. The real test is whether humanity can govern itself in the age of AI.

She noted that around the world, many actors have articulated principles for AI that point in the same direction, but that alignment on shared guardrails, common standards, or practical accountability has not yet been achieved . She explicitly stated that 'principles without practice can inspire false comfort' and that it is not enough to sound responsible while avoiding difficult choices .

Tómasdóttir contends that mere access to AI is insufficient if it does not come with genuine agency for communities. The benefits of AI must reach those who are furthest from today's centres of capital, computing power, and decision-making, and these communities must be co-authors of the AI future rather than passive recipients. Success should be measured by whether people have greater agency, dignity, and hope.

She stated that 'access without agency is not inclusion' and that the benefits of AI must reach communities everywhere, not only those already closest to capital, computing power, and decision-making . She argued that those furthest from power must be co-authors of the future, with a meaningful role in building infrastructure, shaping rules, and sharing fairly in the value created . She proposed measuring success not by the number of pilots announced but by whether people and communities have greater agency, dignity, and hope .

Tómasdóttir argues that as AI becomes more capable, it is essential that meaningful human responsibility is retained in the most consequential decisions. She specifically warns against normalising a future in which machines are delegated the irreversible decision to take a human life. This is framed as both a moral and a governance imperative.

She stated that 'we must not normalize a future in which human beings delegate the irreversible decision to take a human life to a machine' and that 'meaningful human responsibility must remain where the stakes are highest' .

Tómasdóttir calls for a future-fit leadership that brings together a wide range of stakeholders in genuine partnership, rather than leadership that equates technological speed with progress or technical capability with wisdom. She emphasises that young people must play a leading role in shaping the AI future. Iceland's planned national dialogue on AI, in which young people will lead, is offered as a model.

She described Iceland's plan to invite people across the country into a national dialogue in which young people will play a leading role . She called for leadership that 'neither confuses speed with progress nor technical capability with wisdom' and that brings governments, business, educators, researchers, civil society, parents, and young people into genuine partnership .

Tómasdóttir reframes the AI governance challenge as fundamentally a question of human self-governance rather than merely a technical regulatory problem. She draws on Iceland's experience of harnessing powerful natural forces for the common good as an analogy for how AI should be governed. The test for AI is not simply whether it can create value, but whether it can create shared capability, shared opportunity, and shared prosperity without sacrificing human dignity.

She stated that 'the central question is not simply how we govern artificial intelligence, it is whether we can govern ourselves in the age of artificial intelligence' . She drew on Iceland's tradition of turning natural forces to the common good, arguing that the same test applies to AI: not simply whether it can create value, but whether it can create shared capability, shared opportunity, and shared prosperity without sacrificing human dignity, relationships, and sense of purpose .

Smith argues that before AI can be meaningfully deployed globally, the foundational work of completing global electrification must be finished. He presents this as one of several essential steps that must be taken to ensure AI can reach all parts of the world. Without this foundation, AI infrastructure cannot be built equitably.

He stated that 'we still need to complete the electrification of the world' as one of the key steps needed to ensure equitable AI diffusion . He framed this alongside completing internet connectivity, expanding multilingual AI models, and making digital skilling an everyday reality as the kinds of steps needed to 'bend the arc of history' .

Chima argues that the current focus on AI infrastructure and compute deployment is diverting policymakers' attention away from the connectivity needs of communities. He calls for a community-centric approach to internet infrastructure that recognises the importance of connectivity and its impact on climate change. Communities should be treated as a first-mile priority, not a last-mile policy problem.

He stated that 'connectivity efforts are also being impacted by how AI connectivity conversations, sorry, AI infrastructure conversations are grabbing the attention of policymakers and are focusing on particular approaches regarding deployment of compute infrastructure without looking at what communities want' . He called for 'a community-centric approach to Internet infrastructure and one that recognizes the importance of connectivity' and noted the need to connect further with environmental defenders .

Bogantes Zamora argues that AI governance must respect national context, regulatory autonomy, and different levels of institutional capacity rather than imposing a single model. However, connecting these different approaches requires common definitions, shared technical standards, and clear links between legal requirements, technical control, and evidence of compliance. Interoperability must become practical, not merely aspirational.

She stated that 'effective AI governance does not require every country to follow the same model. It requires us to connect different approaches while respecting national context, regulatory autonomy, and different levels of institutional capacity' . She called for interoperability to 'become practical' through 'common definitions, shared technical standards, and clear links between legal requirements, technical control, and evidence of compliance' .

Bogantes Zamora argues that the rapidly evolving nature of AI systems means that static regulatory rules will be inadequate. Governance frameworks must be adaptive to keep pace with systems that are becoming more autonomous, interconnected, and capable of acting across multiple steps. This adaptability is presented as a key principle for effective AI governance.

She stated that 'AI governance must become adaptive. Static rules will not be sufficient for systems that are even more autonomous, interconnected, and capable of acting across multiple steps' .

Bogantes Zamora calls for concrete mechanisms to move AI governance cooperation from principle to implementation. She specifically advocates for cross-border regulatory sandboxes, shared evaluation methods, and incident reporting mechanisms as practical tools that allow countries to test AI systems, learn from failures, and prevent repeated harms. Trust also requires transparency, independent evaluation, and accountability throughout the AI system lifecycle.

She called for 'cooperation to move from principle to implementation' through 'cross-border regulatory sandboxes, shared evaluation methods, and incident reporting mechanisms that can help countries test systems, learn from failures and prevent repeated harms' . She also emphasised that 'trust requires transparency, independent evaluation, and accountability throughout the AI systems' life cycle, especially when AI systems affect access to health care, education, employment, credit, public services, and democratic participation' .

Pröll argues that having listened to governments, scientists, industry, and civil society from every region, the case for a binding framework for AI governance has only grown stronger. He contends that voluntary guidelines are insufficient because good intentions do not stop harm, but rules do. Austria's own experience demonstrates that responsible AI use is possible without slowing innovation.

He stated that 'the world needs a binding framework for the use of AI' and that after listening to all stakeholders, 'that message has only grown stronger' . He argued that 'both are real. Managing both at once is exactly why voluntary guidelines are not enough. Good intentions do not stop harm. Rules do' . He cited Austria's GovGPT platform serving 180,000 public servants, a digital skills initiative reaching thousands of citizens, a minimum age of 14 for social media, and a European Charter on Digital Sovereignty agreed by 27 countries as evidence that responsible AI governance is achievable .

Theofelus argues that a key immediate step for international cooperation is helping countries, particularly those in the global south grappling with structural challenges, to translate global AI principles into effective national laws and institutions. She emphasises that context determines outcomes and that different regions require different approaches. This contextualised translation of principles is presented as a foundational step for any country to handle the AI question.

She called for 'helping countries who are, in many instances, grappling with so many other structural issues, especially us from the global south, to translate those global AI principles into effective national laws and institutions' . She referenced a panel member's statement that 'context determines the outcome' and noted that different regions and contexts mean it is right to condense principles based on each country's context .

Theofelus identifies continuous investment in digital and data infrastructure as a necessary moderate-term approach for developing nations. She acknowledges that some countries are still teaching their citizens basic digital literacy and that pretending all countries are at the same level is not true. Financing must be made available to countries at earlier stages of digital development.

She stated that 'the second practical moderate approach that we can look at is continuously investing in digital and data infrastructure' and acknowledged that 'some countries are still grappling with getting their citizens digitally literate' . She called for finding 'a way to create or avail financing to countries that are still teaching their citizens how to trust technology' .

Theofelus argues that long-term programmes to strengthen skills and scientific capacity in the global south are essential, particularly given Africa's demographic trajectory. She notes that nearly half the world's population will come from the African continent by 2050, bringing both the energy of a young population and significant numbers. Tailor-made training programmes for human capacity are therefore a global necessity.

She stated that 'from a Namibian perspective of a country on the African continent, it's no secret that almost half of the world's population will come from the African continent by the year 2050' . She argued that this means there are 'not only the energy and agility of a young population, but we will have the numbers' and that 'it's only right that, especially for the global south and the developing nations, that there are programs tailor-made to train human capacity and human capabilities for the future of the world' .

Theofelus argues that success at the next global dialogue would include having a shared but flexible reference framework for AI governance that reflects all regions of the world. This framework should give countries like Namibia a practical yet sovereign template for national legislation and institutions. Flexibility is emphasised to accommodate different national contexts.

She stated that success in May 2027 in New York would look like 'having a shared but flexible reference framework for AI governance' that 'reflects on all the regions across the world and giving countries like ours small states, developing states like Namibia where I'm from a practical but sovereign template for national legislation and, of course, our institutions' .

Theofelus argues that visible capacity gains in the global south are a key measure of success, specifically calling for more trained regulators, more national AI strategies deployed, and more cross-regional AI initiatives advancing inclusive growth and resilience. The goal is to ensure that countries do not become dependent on other countries' capabilities for their own AI governance and development.

She stated that success would include 'visible capacity gains in the global south' with 'more trained regulators', 'more national AI strategies deployed', and 'more cross-regional AI initiatives advancing inclusive growth and resilience, especially for countries that are compounded with AI coming up, but they do not necessarily have the capacity in-country, so that there's no dependence on other people's capabilities for one's own country' .

Theofelus argues that global partnerships reflecting Goal 17 of the SDGs are a key measure of success for AI governance. These partnerships should bring together governments, industry, academia, and civil society to regularly exchange evidence, update norms, and co-develop open tools for administering AI. She also emphasises that the voices of all actors must be treated equally.

She called for 'partnerships, global partnerships around this dialogue where governments, industry, academia, civil society are regularly exchanging evidence, updating norms, and co-developing open tools that we can use in administrating the use of AI for now and the future, and that the voices of some are not lesser than others' . She noted that as a government representative she depends on all these actors to properly advise policymakers .

Alshaala identifies the environmental cost of large compute systems as one of several critical emerging blind spots that the AI dialogue must urgently address. She also highlights the threat of synthetic content to information integrity and the need to protect global data centres as critical non-military infrastructure. These blind spots risk being overlooked in the rush to govern AI's more visible impacts.

She called for addressing 'critical emerging blind spots' including 'the environmental cost of the large compute systems, the threat of synthetic content to information integrity, and, crucially, the protection of global data centers, recognizing them as a critical, non-military infrastructure' .

Ryder argues that the UN's unique value in AI governance lies in the legitimacy derived from its universal membership. No other institution can bring together 193 member countries in an all-inclusive conversation that also encompasses business and civil society. This universality gives the UN a particular importance and legitimacy in the AI governance landscape.

He stated that 'the UN enjoys one unparalleled comparative advantage in the task before us, and that is the legitimacy of its universality. There is nowhere else where 193 member countries can come together in this all-inclusive conversation' . He noted that inclusivity encompasses not only all member states but also business and civil society .

Ryder acknowledges that there is a challenge of coherence within the UN system itself, with many different processes and initiatives underway. He argues that the UN must ensure its internal activities are complementary and that it works hand-in-hand across the system. The UN can also serve as a reference point for connecting with other international processes such as the G20 and G7.

He stated that 'we have a challenge, let me be perfectly honest with you, moderator, inside the UN system. There's a lot going on inside the system. We need to make sure that what happens inside the UN system is coherent. It's complementary. We are working hand-in-hand within the system' . He also noted that the UN 'can be a reference point' for other international processes including the G20 and G7 .

Tallinn argues that even the best experts are only now beginning to grasp that AI progress will accelerate from its current pace. He highlights that the independent scientific panel's report makes clear that reliable methods for retaining control over highly autonomous AI systems are lacking. This creates a fundamental governance challenge as the gap between capability and governance widens.

He stated that 'even the best experts are only now starting to grasp that the progress will actually accelerate from here' . He referenced the panel's report, noting that 'reliable methods for retaining control over highly autonomous AI systems are lacking' . He also cited the Secretary General's observation that these systems 'are actually no longer tools awaiting instruction and our intuitions and institutions are not ready for machines that decide' .

Tallinn argues that the top AI companies are now openly pursuing recursive self-improvement, using their own AI to accelerate AI development. He warns that this risks losing control because the autonomy that makes AI systems useful is also what makes them harder to control. In the worst case, humanity could lose control altogether.

He stated that 'the top AI companies are now openly pursuing what's known as self recursive self-improvement that is using their own AI to accelerate the development of AI' . He cited Jack Clark, co-founder of Anthropic, who said their top model set a new world record in a crucial machine learning task . He warned that 'the problem with this recursive self-improvement is that you're risking losing the control because you're giving machines more and more autonomy' .

Tallinn argues that the competitive dynamics driving AI development are pushing companies towards dangerous thresholds even when their own leaders would prefer to slow down. He notes that even leaders of top AI companies are voicing caution and saying they would like to slow down but cannot due to commercial and geopolitical pressure. This is precisely where the UN and international cooperation can step in.

He stated that 'even the leaders of the top AI companies now are voicing their caution. They are saying that they would like to slow down, but they really can't because of this commercial and geopolitical pressure' . He argued that 'this is where UN and more general international cooperation can step in to make sure that there is international pressure to progress at a safe pace' .

Tallinn frames the fundamental governance choice as between deliberate design and passive drifting. He argues that without governance, the default path is to drift in the wind of economic pressure towards dangerous thresholds. However, he emphasises that this is not inevitable and that humanity can choose a better path.

He stated that 'the choice, as the secretary general puts it is between governing by design and drifting and that's not inevitable by default probably drifting in the wind of economic pressure. But in the end, it is the path that we, humanity, are choosing, and we can choose a better path' .

Tallinn argues that urgency is the most important message to carry forward from the dialogue. He warns that AI developments are now happening on a quarterly basis and will soon be monthly and then weekly. If governance continues to operate on a yearly basis, there will be a fundamental mismatch that prevents effective oversight.

He stated that 'urgency' is what he hopes is carried forward, 'because I think the situation in AI is now on an exponential where the developments certainly happen on a quarterly basis. Soon they will be monthly and then weekly. And if the governance continues to work on a yearly basis, there's a problem' .

Tallinn argues that the apparent tension between innovation and safety is a false dilemma. He identifies two distinct frontiers: the capability frontier where companies race into the unknown (where safety concerns lie) and the adoption and diffusion frontier where communities tap into existing capabilities (where innovation benefits come from). Addressing safety concerns in the first frontier does not impede the innovation benefits of the second.

He described 'two frontiers that are being pushed': one making AI more capable, happening behind gates controlled by AI companies and host countries, and another around AI adoption and diffusion . He concluded that 'there is really no dilemma anymore between innovation and safety because most of the safety concerns are in the first frontier where the companies are racing into unknown and most of the innovation and benefits actually come from the second frontier where the companies and the community in general can actually tap into those capabilities' .

Tallinn argues that the UN, as the most legitimate organisation on the planet, is uniquely placed to build the international cooperation needed to counter the competitive and geopolitical pressures driving unsafe AI development. International pressure through the UN can help ensure that AI progresses at a safe pace so that humanity can capture the benefits of increasing capabilities without succumbing to loss of control.

He argued that 'this is where UN and more general international cooperation can step in to make sure that there is international pressure to progress at a safe pace so we can actually capture the AI benefits' . He stated that 'with this dialogue and the scientific panels independent evidence the UN obviously has the most legitimate organization on the planet is uniquely placed to build that cooperation' .

Marcus argues that trust in AI governance cannot be treated as a side event or consultation exercise but must be embedded in the governance process itself. She identifies trust as being built through specific qualities of the governance process: reliability, responsiveness, integrity, fairness, openness, and above all accountability. These are not outcomes to be asserted but qualities to be demonstrated.

She stated that 'embedding trust and inclusion into this dialogue cannot mean treating participation as a side event or a consultation exercise. Trust is built through reliability, responsiveness, integrity, fairness, openness and above all, accountability. And these are qualities of the governance process itself' .

Marcus argues that global AI governance should be shaped by a shared understanding of evidence that includes not only technical evidence about AI systems but also evidence about how impacts are being experienced by real people. She contends that those most affected by AI technologies, including workers and vulnerable communities, are often the true experts in their impacts. Scientific evidence is essential but often lags behind the pace of change.

She stated that 'we need to treat public participation as a source of evidence' and that governance should be shaped by 'a shared understanding of evidence, not only the technical evidence about AI systems, but evidence about how impacts are being experienced by real people across the world' . She noted that 'it is often the people most affected by these technologies that are the true experts in their impacts' and gave examples of workers understanding automation effects and vulnerable groups bringing lived experience of technologies deployed before they are governed .

Marcus argues that diverse mechanisms of participation, whether citizen tracks, national and regional dialogues, or participatory incident reporting mechanisms, must be connected to clear levers of power and institutional accountability. The UN dialogue must be the place where these mechanisms contribute to outcomes with clear accountability. Participation must mean shaping agendas, evidence, priorities, and recommendations, not merely collecting views.

She stated that 'the dialogue must be the place where different mechanisms, be they a citizen's track of affected communities, national and regional dialogues, or participatory incident reporting mechanisms, the dialogue must be the place where these mechanisms contribute to outcomes that have clear accountability' . She emphasised that 'participation does not just mean collecting representing views, but making sure that those views shape agendas, evidence, priorities and recommendations' .

Marcus argues that inclusion of marginalised people and those with the most to gain from AI will not automatically lead to trust or change. Their experiences and expectations must be connected to clear levers of power for inclusion to be meaningful. This requires an embedded and institutionalised participation architecture with clear accountability.

She stated that 'inclusion of marginalized people and of those who have the most to gain from AI will not automatically lead to trust and it will not lead to change unless their experiences and expectations are connected to clear levers of power' . She called for 'developing an embedded and institutionalised participation architecture around the dialogue that adds substance to a recognised, respected and defended global governance floor' .

Finlay argues that while the independent scientific panel's report is laudable and crucial, it should be understood as providing the floor of concern rather than the ceiling. Scientific evidence in AI often lags behind the pace of change, meaning the panel's findings represent a minimum baseline of concern rather than a comprehensive account of all risks. This underscores the need to continuously strengthen the evidence base.

She stated that 'the independent scientific panel gives us the floor of our concern not the ceiling and while our focus must be on human rights protecting safe and trusted ai for all it is often the people most affected by these technologies that are the true experts in their impacts' . She called for supporting the panel and cohering the body of evidence already gathered, linking it between the UN, State of Safety Reports globally, and the International Network of AI Safety Institutes .

Finlay argues that the UN dialogue has the potential to be the place where coherence is achieved around open benchmarks, verification, independent testing, and technical standards. She notes that global commitments are already in place through various processes, but these need to be opened up with more open benchmarks and independent testing. The dialogue could ensure that models reflect linguistic diversity and that open science and technical standards are advanced.

She referenced existing global commitments including the Hiroshima process, international law, human rights frameworks, and pledges from AI summits . She stated that 'the UN dialogue could be the place where that coherence happens, and we could dig into some of those challenges that are required for full equity and inclusion' through 'open benchmarks, verification, independent testing, making sure that models reflect linguistic diversity, open science, and technical standards' .

Doumba argues that the real frontier of AI opportunity lies not in building bigger models but in local adaptation that meets people where they are. He contends that AI today is shaped by Western infrastructure, language, and values, but the cognitive economy only works if it is adapted to local contexts. The disproportionate value for economies whose knowledge has always been tacit rather than codified lies in this local adaptation.

He stated that 'most of AI's opportunity will come from adapting big AI to small AI. AI today is shaped by Western infrastructure, language and values but the cognitive economy only works if we meet people where they are and that means the frontier isn't getting bigger model. It's about local adaptation that lets an African former or a Pacific Island utility operator actually use this technology in their own context' .

Doumba argues that the instinct to compete on speed in AI deployment misses the fundamental point. The real contest is about who builds the institutions and judgement to deploy AI well, not who deploys it fastest. This reframing challenges the prevailing competitive dynamic and calls for a focus on quality of governance over speed of deployment.

He stated that 'the first is to not race to be first on AI. Race to use AI wisely. The instinct to compete on speed misses the point. The real contest is who builds the institutions and judgment to deploy AI well as opposed to who deploys it fastest even if this can feel counterintuitive in the world that we live in' .

Doumba argues that AI is not business as usual and does not simply ask us to optimise existing systems. It demands a fundamental redesign of systems with new incentives, new measures of success, and new assumptions about who gets to participate and how. He frames the current moment as one of unprecedented capital, talent, and technology that must be used to distribute abundance fairly.

He stated that 'AI is a paradigm shift. It's not business as usual. It doesn't ask us to optimize existing systems. It demands that we redesign them with new incentives, new measures of success, new assumptions about who gets to participate and how' . He noted that 'there has never been more capital, talent, technology assembled than today' and posed the question of whether this abundance will be distributed fairly or whether the gap between who builds AI and who benefits will become 'this century's defining inequality' .

Khan argues that the outcomes of AI deployment must be measured honestly, and this measurement must include AI's own environmental footprint across both training and inference. He frames this as part of a broader commitment to ensuring that AI serves genuine human needs rather than becoming an end in itself. Honest measurement is presented as a governance imperative.

He stated that 'outcomes are the mandate for all of us today, and those outcomes must be measured honestly, including AI's own footprint across training and inference' .

Diouf argues that the digital divide is not limited to AI but extends across infrastructure, connectivity, devices, and knowledge. For countries like those in West Africa, the challenges are at the most fundamental levels of connectivity and infrastructure, not yet at the level of AI. Bridging the gap requires addressing all these layers together through collective action and knowledge sharing.

He stated that 'this divide, we applied on the AI, but it goes beyond that. We applied on the infrastructure. We applied on the connectivity. We applied on the device. We applied on the knowledge' . He noted that for West Africa, 'we are on the connectivity level we are on the infrastructure level we are on the devices level and to go beyond the AI we need to fix those challenges' .

Kurbalija presents a taxonomy of digital inclusion gaps identified during the cluster discussion, spanning access (infrastructure), funding, participation, skills, knowledge, language, and trust. He notes that knowledge was mentioned 35 times during the discussion compared to data 63 times, suggesting knowledge may be more important than data in this context. Local creation, co-creation, and local expertise are identified as key elements around knowledge.

He outlined a taxonomy of digital inclusion gaps including 'access' (infrastructure), 'funding gap', 'participation gap', 'skills gap', 'knowledge' (mentioned 35 times versus data 63 times), 'language', and 'trust' . He noted that around knowledge, key words included 'local creation, co-creation, and local expertise' and emphasised 'context, context, context' .

Bonyo argues that the evidence from the scientific panel confirms that AI governance globally is being run by a few states where power is concentrated. She identifies visa refusal rates as a concrete mechanism that perpetuates this imbalance by preventing participation from the global south. She calls for transparency and accountability in the AI systems used to determine visa refusals.

She stated that 'we agree with the evidence from the scientific panel that AI governance globally is being run by a few states. The power is concentrated there' . She noted that 'the last six months, 51% of conversations on AI governance happened in Geneva' and that 'Africa accounts for 7 out of 10 visa refusal rates' . She cited a specific example of an AI worker from the Commonwealth who could not attend due to visa refusals .

Bonyo highlights visa refusal rates as a concrete barrier to meaningful participation from the global south in AI governance discussions. She notes that Africa accounts for 7 out of 10 visa refusal rates and that the average African participant has spent 2,500 euros to attend, compared to 200 euros for some others. She calls for transparency and accountability in the AI systems used to determine visa refusals and proposes an AI passport to facilitate movement.

She stated that 'the visa refusal rates for Africa, Africa accounts for 7 out of 10 visa refusal rates' . She noted that 'the average African has spent 2500 euros to get to this place, some of you spent only 200' . She called for 'transparency and accountability for the AI systems that are being used to determine visa refusals' and proposed 'an AI passport' to make it easier for people to move when they want to invest in AI or participate in AI governance .

Bonyo argues that civil society organisations are increasingly absent from AI governance discussions because they lack funding. She notes that there are fewer civil society organisations present at the current dialogue than previously. New and disruptive ways of financing human rights in the digital age must be found to ensure civil society can continue to participate meaningfully.

She stated that 'many civil society organizations or organizations just do not have the funding anymore. In this room, there are fewer civil society organizations this time. We must look at disruptive and new ways to do that' .

Bonyo argues that the voices of AI workers are missing from AI governance conversations. She highlights the ILO's recent convention on decent work and calls on states to support and ratify these conventions to address the harms of algorithmic management. She also raises concerns about the general sentiment around algorithmic management and the need to domesticate and implement global principles at the national level.

She stated that 'we talked about AI workers and who's missing in the room. We do not have voices from the AI workers' . She referenced the ILO convention, noting that 'ILO just passed, you know, I think 193, you need to read it up on decent work for these AI workers that are crying every day on algorithmic management. It is time that states support and ratify these conventions' . She also noted that Lawyers have launched an AI governance index for Africa showing that 'states will sign these treaties but do not domesticate them in their country' .

Rossbach argues that cities and local governments are on the front lines of AI application, making decisions that directly affect people's lives and generating crucial evidence for AI policymaking. Despite this, they remain largely absent from global AI governance deliberations. She warns that if data gaps in informal settlements and vulnerable communities are not addressed, AI risks reinforcing existing inequalities.

She stated that 'cities and local governments are where artificial intelligence is already being applied every day. They make decisions that directly affect people's lives and generate crucial evidence for AI policymaking. Yet, despite this reality, they remain largely absent from global AI governance deliberations' . She noted that 'in many cities, people living in informal settlements and other vulnerable communities remain isolated. They are underrepresented or entirely absent from official data sets' and that if these gaps are not addressed, 'AI risks reinforcing existing inequalities' .

Popescu argues that the protection of children from the harms of AI is one of the most important issues in AI governance. He notes that AI increasingly shapes what children see, learn, and are exposed to online. He frames child protection not merely as a policy preference but as a matter of principle.

He stated that 'AI increasingly shapes what children see, learn, and are exposed to online, and we need to protect them above. This is a matter of principle, not only of presence' .

Popescu argues that the transition to an AI-shaped economy must be deliberately managed to avoid leaving the most vulnerable behind. He contends that progress that concentrates rewards while distributing only disruptions is not progress but instability postponed. A deliberate transition architecture through retraining and pathways into new jobs is therefore needed.

He stated that 'the move to an economy shaped by AI must be not allowed to leave the most vulnerable behind. Progress that concentrates its rewards and distributes only its disruptions is not progress. It's instability postponed. We therefore need a deliberate transition architecture through retraining and pathways into the new jobs this technology creates' . He noted that this is not a task for governments alone and saluted private actors investing in free, cross-sector training credentials .

Lincolao argues that as compute becomes a key driver of the new AI industry, it must be accompanied by renewable energies and technical capacities. She frames AI not only as a technology but as a critical infrastructure requiring systemic vision that encompasses digital, physical, energy, and industrial dimensions. Investment in talent and accountability are also highlighted as necessary.

She stated that 'AI is not only a technology, but also a critical infrastructure. Therefore, its governance requires systemic vision, which includes digital and the physical. Technology is also policy and energy and around industry as well' . She noted that 'these days has shown us that compute is one of the drivers for the new industry' and that 'there has to be responsibility to have renewable energies that are abundant and technical capacities which allow us to have an ideal society for this new stage' .

Bdeir argues that all over the world there are thousands of builders, governments, nonprofits, and companies creating open-source AI building blocks, including datasets, foundation models, evaluation toolkits, and benchmark systems. The problem is that these components are not stitched together and are not interoperable. Current AI's mission is to invest in the connective tissue between these tools to create an end-to-end, vertically integrated, open AI stack.

She stated that 'all over the world, there are thousands and thousands of builders, of governments, of nonprofits, of companies that are creating building blocks for an open source AI stack. Data sets, foundation models, evaluation toolkits, benchmark systems. The problem is that they are not stitched together and they are not interoperable. So that is what we are investing in' . She noted that Current AI has 'surveyed over 24,000 of these tools and components all over the world' and identified the gaps .

Bdeir draws on historical precedents to argue that transformative technologies can be built through international technical collaboration rather than by a single actor or nation. She cites the World Wide Web and Linux as examples of technologies built through international collaboration that now underpin global digital infrastructure. These models offer a template for building public-interest AI.

She stated that 'we already have models of international technical collaboration by looking at history. The world wide web, the other transformative technology of modern times was built through an international technical collaboration. Linux, which powers 95% of servers in the world was built through international technical collaboration' .

Bdeir argues that the speed at which AI is developing means that traditional forms of international collaboration, such as MOUs and legal agreements, are insufficient. Open code bases offer the ability to collaborate at the speed of AI. She illustrates this with the example of a chatbot whose components come from Switzerland, Finland, India, and US nonprofits, stitched together through open collaboration.

She stated that 'this requirement of international technical collaboration that AI demands of us right now cannot happen through traditional collaboration through MOUs and lawyers. We have to collaborate at the speed of AI and open code bases offer us this ability' . She described a chatbot being launched where 'the foundation model is from Switzerland, the compute comes from Finland, the data sets are coming from India, the safety safeguards are coming from a non-profit in the US' .

Chatardová argues that a human-centric approach focusing on the needs and perspectives of human beings must remain at the heart of AI development and use. She emphasises that people and their rights must remain central throughout the entire lifecycle of AI systems. This is presented alongside the need for transparency, accountability, and meaningful human oversight to build public trust.

She stated that 'a human-centric approach focusing on the needs and perspectives of human beings must remain at the heart of AI development and use, ensuring that people and their rights remain central throughout the lifecycle of AI systems' . She also noted that 'public trust in AI depends on transparency, accountability and meaningful human oversight' .

Chatardová argues that the United Nations has an important role as a universal platform for AI governance, bringing together the full range of stakeholders. She also emphasises that companies developing and deploying AI systems bear a key responsibility for ensuring these technologies are used safely and responsibly. This multi-stakeholder approach is presented as essential for effective AI governance.

She stated that 'AI governance requires global cooperation. The United Nations has an important role as a universal platform, bringing together governments, international organizations, the private sector, academia, the technical community, and civil society' . She also noted that 'companies developing and deploying AI systems also bear a key responsibility for ensuring that these technologies are used safely and responsibly' .

Chatardová argues that different countries and regions will pursue different AI approaches, but the focus should be on interoperability rather than fragmentation. She calls for working towards shared principles and compatibility that facilitate international cooperation, innovation, research, and trade, including through technical standards. Building advanced computing capabilities and research capacity should also be integral to the global AI agenda.

She stated that 'we should focus on interoperability rather than fragmentation. Different countries and regions will pursue different approaches, but we should work towards shared principles and compatibility that facilitate international cooperation, innovation, research and trade, including through technical standards' . She also called for 'building these capacities' including 'talent, research capacity, and expand advanced computing capabilities' as 'an integral part of the global AI agenda' .

Chatardová argues that realising AI's potential for innovation, economic growth, and societal resilience depends not only on the rules set but also on the ability to develop talent, research capacity, and advanced computing capabilities. Building these capacities should therefore be an integral part of the global AI agenda, not treated as separate from governance discussions.

She stated that 'realizing this potential will depend not only on the rules we set, but also on our ability to develop talent, research capacity, and expand advanced computing capabilities. Building these capacities should therefore be an integral part of the global AI agenda' . She also noted the Czech Republic's contribution through the Czech AI Factory, 'part of a growing European network supporting AI research, innovation, and access to advanced computing capabilities' .

Chatardová argues that AI governance must remain inclusive and multi-stakeholder in nature. She contends that public trust in AI depends on transparency, accountability, and meaningful human oversight. These are presented as foundational requirements for legitimate AI governance.

She stated that 'AI governance must remain inclusive and multi-stakeholder. Public trust in AI depends on transparency, accountability and meaningful human oversight' .

Borisenko argues that access to frontier AI technologies is not universal, with key players in models, cloud, and chips sometimes applying restrictions that give some exclusive access while others face obstacles. Russia presents itself as ready to share its AI experience and models with other countries while respecting their cultural codes and specificities. He frames this as contributing to common standards that reduce the risk of AI becoming the privilege of a small circle.

He stated that Russia has 'a full technical chain to create advanced models without needing to use any foreign components' including two large language models GigaChat and Yandex GPT . He stated that Russia is 'ready to share our experience with other countries to respect their cultural codes and cultural specificities' and that 'we will not be scraping and not be using this data to for our benefit' . He argued that 'common initiatives gives access to instruments to open models and to shape common standards' and that 'this reduces the risk that AI will be the privilege of a small circle of people' .

Ndiaye argues that the UN-supported global network of centres for exchange and cooperation on AI capacity building demonstrates a practical model for international cooperation. Rather than creating another institution, the network connects existing centres of excellence into a collaborative network where countries can exchange expertise, share knowledge, develop talent, and support one another. This approach has already produced a cooperation framework in just a few months.

He stated that 'what makes this initiative unique is that it is not creating another institution. It is connecting existing centers of excellence into a collaborative network where countries can exchange expertise, share knowledge, develop talent, strengthen national AI ecosystems, and support one another according to their respective strengths' . He noted that 'in only a few months, participating centers have jointly developed a cooperation framework that defines common principles, governance arrangements, and priority areas for collaboration' .

Ndiaye argues that global AI capacity building cannot be delivered through funding alone. It also requires trusted partnerships, peer learning, and long-term institutional cooperation. The network is designed as a platform where governments, universities, research institutions, the private sector, and philanthropic organisations can combine their comparative advantages to accelerate AI capacity worldwide.

He stated that 'global AI capacity building cannot be delivered through funding alone. It also requires trusted partnerships, peer learning, and long-term institutional cooperation' . He described the network as 'designed as a platform where governments, universities, research institutions, the private sector, and philanthropic organizations can combine the comparative advantages to accelerate AI capacity worldwide' .

Tammsaar, in his role as co-chair, emphasises that the dialogue was convened by a decision of all 193 UN member states, which gives it particular importance and legitimacy. He frames this as a reminder of both the content of the mandate and who it comes from. He also acknowledges the extraordinary contributions made to make the dialogue work successfully.

Guy Ryder, speaking in the same session, noted that 'it was only in August of last year that our member states decided to convene this dialogue. This dialogue is happening because 193 countries of the world decided that it would be a good idea to come together and have this dialogue and it's important to remember the mandate not only what its content is but who it comes from and it gives a particular importance and a particular legitimacy to what we're all doing here together today' .

Melamed, as moderator, frames the purpose of the concluding session as drawing together the many perspectives shared throughout the dialogue and contributing to a stronger understanding of where further cooperation, coordination, and collective action may be most valuable. She emphasises that the session should help chart the way forward and identify key messages emerging across the dialogues. The goal is to set the stage for the second global dialogue in New York.

She stated that the session 'will provide an opportunity to identify key messages emerging across the dialogues, consider areas of convergence and complementarity, and explore how different initiatives, experiences, and approaches can inform and reinforce one another' . She described the dialogue of dialogues as beginning 'to chart the way forward and think about how the UN can play a key role in ensuring that AI governance is built upon the foundations of human rights and of the values espoused by the UN' .

López argues that the ideas and experiences shared over the two days of dialogue underscore that effective AI governance will require sustained collaboration beyond the event itself. She frames the dialogue not as a conclusion but as a starting point, with all participants having a role in carrying its messages forward to the next global dialogue. This positions the dialogue as one step in an ongoing process rather than a self-contained exercise.

She stated that 'the conversation on AI governance cannot end with this dialogue. And the experiences, the ideas shared over the past two days underscore the advanced effective AI governance will require continued collaboration across sectors, regions, and initiatives' . She further noted that 'we all have a role to play in carrying these critical messages forward to the next global dialogue' .

López argues that the reporting back from thematic cluster co-chairs and the subsequent interventions from member states and stakeholders are a necessary part of the dialogue process, ensuring that the full range of perspectives is captured. She emphasises fairness and inclusivity in participation, noting that some participants had not had the opportunity to speak during the two days. This reflects a commitment to ensuring that the dialogue's outcomes reflect diverse voices.

She explained the rationale for the reporting-back process, stating that 'we have the possibility to hear from you and we're going to be able to hear from you hearing back from member states and other stakeholders also on those discussions and some of the people that were going to listen after that, they were not able to speak throughout these two days' . She also noted that the mic would be automatically cut off to ensure fairness and allow for broad participation .

López, together with her co-chair, emphasises that written submissions are an important mechanism for capturing the full range of perspectives from participants who may not have had the opportunity to speak. This reflects a commitment to inclusivity and to ensuring that the dialogue's record is as comprehensive as possible. The invitation to submit statements to the secretariat is framed as a way of ensuring that no voice is lost.

Her co-chair Rein Tammsaar, speaking immediately after her handover, relayed the request that participants 'please do not forget to send your statement, valuable statements to the email address AI dialogue at UN.org' , which was part of the procedural guidance that López had set up and handed over to him to deliver.