Harvard Kennedy School and 10Billion.org

A successful first Global Dialogue on AI Governance should produce three things. First, a clear and credible roadmap for future work. The Dialogue should not end as a one-off exchange of views. It should identify priority areas for continued cooperation, clarify where further technical work is needed, and show how consultations, written inputs, and future sessions in Geneva and New York will build on one another. Second, practical outputs that are useful across very different national contexts. Many countries need governance options that are modular, proportionate, and feasible even without frontier AI developers or large regulatory capacity. Success would mean surfacing concrete policy building blocks, good practices, and capacity-support needs rather than only reaffirming high-level principles. Third, meaningful inclusion. The Dialogue should help ensure that developing countries, civil society, technical experts, and underrepresented linguistic and regional communities are shaping agendas early, not merely reacting to decisions made elsewhere. Bridging AI divides should include voice, institutional capacity, and access to evaluation and oversight tools, not only access to technology. More broadly, the first Dialogue would be successful if it strengthens trust that the UN can convene a serious, inclusive, and implementation-oriented process on AI governance. The best outcome would be broad legitimacy combined with enough specificity to improve policy design, coordination, and accountability across jurisdictions.

• Safe, secure and trustworthy AI
• AI capacity-building
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These four priorities are the most urgent because they address both the legitimacy and the practical effectiveness of AI governance. Safe, secure and trustworthy AI is essential because international cooperation will fail if governance processes do not meaningfully address misuse, systemic risks, and the reliability of high-impact AI systems. AI capacity-building is equally important because many countries and communities still lack the technical, institutional, and regulatory capacity needed to evaluate systems, participate in governance, and implement safeguards. Without stronger capacity, global AI governance risks becoming governance by a few for the rest. Protection and promotion of human rights must remain central because AI systems increasingly affect privacy, expression, non-discrimination, due process, labor, and access to essential services. Human rights should not be treated as a side issue. They should shape the overall architecture of governance. Transparency, accountability, and human oversight matter because principles alone are not enough. Governance needs practices that can be monitored and improved over time, such as clearer reporting, documentation, redress mechanisms, and proportionate oversight for higher-risk uses. Together, these four areas can help move the Dialogue toward governance that is globally relevant, operationally credible, and more inclusive of countries and communities that risk otherwise becoming rule-takers rather than rule-shapers.

Yes. One important cross-cutting issue is implementation capacity and institutional asymmetry. Many governance debates assume a level of administrative, technical, and financial capacity that much of the world does not yet have. This shapes whether transparency rules, safety practices, or human rights protections can actually be enforced. A second issue is the concentration of compute, data, and model development power. AI governance is not only about downstream use. It is also about upstream concentration in infrastructure, talent, and supply chains. These structural asymmetries affect competition, dependency, bargaining power, and the ability of states to govern effectively. A third issue is evaluation and measurement. International dialogue will be more useful if there is greater attention to shared terminology, comparable evidence, and governance metrics that allow countries and institutions to learn from one another without requiring full legal harmonization. A fourth issue is policy continuity across jurisdictions and over time. Many stakeholders are overwhelmed by fragmented initiatives, overlapping principles, and unclear follow-up. The Dialogue could add value by improving coherence between existing international, regional, and national processes. Finally, the governance implications of increasingly general-purpose and agentic AI systems deserve attention. These systems may blur boundaries between sectors and create new challenges for responsibility, oversight, and incident response. That makes it important to design governance approaches that remain adaptable as capabilities evolve.

The governance gaps in these areas are already shaping both the opportunities and the risks facing civil society actors working across Europe and globally. The most significant challenge is asymmetry. A small number of firms and jurisdictions are moving quickly on frontier AI, while many governments, researchers, and civil society organizations still lack the capacity to evaluate systems, influence standards, or implement safeguards. This creates a risk that much of the world becomes a rule-taker rather than a rule-shaper. A second challenge is fragmentation. Different approaches to safety, transparency, human rights, and accountability are emerging across jurisdictions, but they are not yet sufficiently interoperable. For organizations working internationally, this increases uncertainty, raises compliance and coordination costs, and makes it harder to know which practices should be treated as baseline expectations. A third challenge is implementation. High-level principles are becoming more common, but practical mechanisms for oversight, redress, incident reporting, and independent evaluation remain uneven. This weakens trust and makes it harder to distinguish responsible deployment from mere claims of responsibility. At the same time, there are important opportunities. AI can expand access to knowledge, translation, education, and public-interest services across borders. It can also help smaller organizations work more effectively and reach more people. If governance frameworks become more inclusive and actionable, they could lower barriers to beneficial adoption while reducing harms. For civil society, the central opportunity is to help shape governance that is globally relevant, rights-respecting, and feasible beyond the most advanced economies. That would improve legitimacy, reduce governance gaps, and make international cooperation more credible.

The AI Dialogue can play a uniquely valuable role as an inclusive UN platform that connects governments, international organizations, civil society, academia, and the technical community around a shared governance agenda. The General Assembly created it precisely as a forum for international cooperation, exchange of best practices, and open, transparent, and inclusive discussion, so its comparative advantage is broad legitimacy and convening power. Its most useful role is to reduce fragmentation. Today, AI governance is developing across many national, regional, and plurilateral tracks. The Dialogue can help map where approaches are converging, where important differences remain, and where interoperability is both feasible and desirable. It can also elevate perspectives from countries and communities that are often underrepresented in standard-setting debates. The Dialogue can add real value if it becomes a bridge between broad principles and practical cooperation. That means identifying concrete areas for follow-up such as capacity-building, common terminology, policy learning, incident reporting practices, evaluation approaches, and human-rights-respecting governance measures for higher-impact systems. It can also help create continuity across initiatives. Rather than duplicating existing work, the Dialogue should connect it, highlight lessons that travel across regions, and make it easier for stakeholders to see how different frameworks relate to one another. In short, the Dialogue can advance international cooperation by providing legitimacy, inclusion, and policy coherence. If it succeeds, it can become a place where diverse stakeholders shape a more interoperable, implementation-oriented, and globally relevant approach to AI governance.

The AI Dialogue should build on existing international efforts rather than start from zero. Important reference points include the OECD AI Principles and OECD.AI work on policy metrics and implementation, UNESCO's Recommendation on the Ethics of Artificial Intelligence, and the Council of Europe Framework Convention on AI, Human Rights, Democracy and the Rule of Law. These frameworks already provide widely recognized foundations on trustworthy AI, human rights, accountability, and governance practice. It should also connect with more recent plurilateral and expert processes, including the G7 Hiroshima AI Process and the recommendations of the UN High-level Advisory Body on AI. The latter explicitly pointed to existing initiatives such as the OECD AI Principles, the G7 Hiroshima AI Process, and the Council of Europe Convention as part of the broader governance landscape the UN should build upon. The added value of the AI Dialogue is different from these initiatives. It is not mainly to create another standalone standard. Its value is to provide a more universal and inclusive forum under UN auspices where governments and non-state stakeholders from all regions can compare approaches, surface gaps, and identify pathways for cooperation that are relevant beyond a limited club of states. In practice, the Dialogue could add value by improving coherence across existing frameworks, amplifying voices from underrepresented regions, and translating broad principles into more usable governance options for countries with different levels of technical and institutional capacity. That would make the wider ecosystem more connected, more legitimate, and more actionable.

Different stakeholders should contribute in ways that reflect their comparative advantages. Governments can share regulatory approaches, implementation challenges, and priorities for cooperation. International organizations can help connect existing frameworks and identify areas for alignment. Civil society can surface rights, inclusion, and accountability concerns that may otherwise be overlooked. Technical experts and researchers can clarify what is feasible, what remains uncertain, and where governance tools need to be updated. Private sector actors can provide information on deployment realities, risk management practices, and operational constraints. To make this effective, the Dialogue should combine plenary discussion with more structured thematic work. High-level sessions are useful for legitimacy and agenda-setting, but they should be complemented by smaller, well-moderated discussions focused on specific questions, trade-offs, and governance options. It would help if each thematic cluster had a short background note, a limited number of guiding questions, and a mechanism for synthesizing stakeholder input into concise outcomes. Written submissions should be summarized transparently so participants can see how inputs are reflected. The Dialogue should also provide continuity across meetings by showing what was heard, what remains contested, and what will be carried forward. Hybrid participation is important, but inclusion requires more than streaming. Time-zone sensitivity, multilingual access, concise briefing materials, and support for participants with fewer resources would improve the quality of engagement. A successful structure would combine openness with enough discipline to move from broad discussion toward practical cooperation.

Several voices remain underrepresented in global AI governance discussions. First, many stakeholders from developing countries still have too little influence over agenda-setting, even when they are invited to comment. This includes policymakers from lower-capacity states, local civil society organizations, public-interest technologists, and researchers working outside the most visible transatlantic institutions. Second, linguistic diversity remains limited. Global debates are still disproportionately shaped in English, which narrows participation and can exclude important regional expertise, social concerns, and policy perspectives. Third, workers and communities most affected by AI deployment are often less visible than those building or regulating the technology. This includes educators, journalists, artists, translators, gig workers, data workers, and communities affected by automated decision-making in public and private services. Fourth, smaller enterprises and non-frontier technical communities are often overshadowed by a few major firms and leading labs, even though governance choices will affect innovation ecosystems much more broadly. Inclusion should therefore be designed, not assumed. This means earlier outreach, multilingual calls for input, financial and logistical support where needed, more balanced speaker selection, and stronger pathways for written submissions to shape agendas. It also means creating formats where less-resourced participants can engage substantively rather than symbolically. The goal should be to ensure that those who are most affected by AI governance, and those who otherwise risk becoming rule-takers, can help shape the terms of debate from the beginning.

A mix of formats would likely work best. First, the Dialogue could use short, problem-focused roundtables instead of relying mainly on long prepared statements. Small moderated sessions built around a specific question or governance dilemma would encourage more substantive exchange and make it easier to compare perspectives across sectors and regions. Second, scenario-based discussions could be especially useful. For example, participants could react to a hypothetical cross-border AI incident, a dispute over model transparency, or a capacity gap in a lower-resource context. This would make abstract principles more concrete and reveal where governance approaches differ in practice. Third, structured response formats could improve quality. Participants might be asked to identify one priority, one practical obstacle, and one recommendation. This would make interventions more concise and easier to synthesize. Fourth, the Dialogue could include stakeholder clinics or implementation workshops where participants exchange concrete tools, examples, and lessons learned rather than only high-level views. These could be particularly valuable on topics such as incident reporting, evaluation, documentation, and human oversight. Finally, digital participation should be designed for real engagement. This could include asynchronous written comment windows before and after sessions, multilingual summaries, and visible synthesis of key inputs so that participants can see how their contributions shaped the discussion. The most effective format will be one that combines openness, interaction, and disciplined synthesis so the Dialogue becomes both inclusive and decision-useful.

Examples of useful approaches already exist at different levels. At the international level, the OECD AI Principles provide a flexible foundation for trustworthy AI that respects human rights and democratic values, while the OECD.AI ecosystem helps translate those principles into policy learning, metrics, and implementation support. UNESCO's Recommendation on the Ethics of Artificial Intelligence is valuable because it is global in scope and explicitly connects AI governance to human rights, human dignity, transparency, accountability, and human oversight. The Council of Europe Framework Convention on AI, Human Rights, Democracy and the Rule of Law is another important example because it shows how governance can be anchored in legal safeguards and procedural protections across the AI lifecycle. At the operational level, the NIST AI Risk Management Framework is a strong example of a practical tool that organizations can use to identify, assess, and manage AI risks in a structured way. Its emphasis on governance, measurement, and ongoing risk management is especially useful for implementation. Across these examples, a few common strengths stand out. Effective approaches are risk-based, adaptable across contexts, grounded in rights and accountability, and supported by practical guidance rather than principle alone. They also recognize that governance is not only about restricting harm, but also about building institutional capacity and trust. The AI Dialogue could add value by connecting these existing approaches more clearly, identifying where they are complementary, and helping adapt them for countries and organizations with different levels of technical and regulatory capacity. That would make the broader AI governance landscape more coherent, more inclusive, and more usable in practice.