Self

A successful first Dialogue should leave Geneva with a small, practical cooperation package, not just a broad exchange of principles. In my view, five outcomes matter most. First, a shared crosswalk of major AI governance approaches, so countries can compare and translate between frameworks instead of starting from zero. Second, a light-touch assurance packet for high-impact AI uses, covering intended use, testing, human oversight, incident reporting, and update or rollback arrangements. Third, a capacity compact that matches countries needing compute access, skills, evaluation tools, and local-language resources with partners able to provide them. Fourth, a common incident-learning channel, so countries and organizations can share lessons from failures, near misses, and corrective actions. Fifth, a clear follow-up path to 2027, through thematic tracks, practical partnerships, and a brief public progress update. This would fit the Dialogue's design. The UN has made clear that this is not a negotiating forum and that its value lies in building shared understanding over time. The draft note also emphasizes coherence, interoperability, and practical policy insights for the Co-Chairs' summary. A focused package of tools, templates, and follow-up mechanisms would therefore complement the Dialogue's mandate better than trying to settle every contested issue in one meeting. Success is not agreement on everything. Success is leaving with a common vocabulary, a short list of usable tools, and a credible workplan that helps every country participate more effectively.

• Safe, secure and trustworthy AI
• AI capacity-building
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

I selected safe, secure and trustworthy AI; AI capacity-building; interoperability of governance approaches; and transparency, accountability, and human oversight because together they form the minimum operating layer for credible global cooperation. Safe, secure and trustworthy AI is the entry condition for adoption. If people and institutions do not trust systems in practice, benefits will not scale. Capacity-building is the fairness condition. AI opportunity is still unevenly distributed: ITU reports 2.2 billion people remain offline, and the World Bank finds low- and middle-income countries face steep challenges in compute, infrastructure, skills, and local adaptation. Interoperability is the efficiency condition. Countries are building rules through different legal and institutional traditions; the UN Dialogue can add value by helping these approaches work across borders rather than fragmenting into incompatible regimes. Transparency, accountability, and human oversight are the legitimacy condition. People need to know when AI is used, who is responsible, how decisions can be challenged, and when humans can intervene. UNESCO's Recommendation and ISO/IEC 42001 both support this lifecycle-based, governance-oriented approach. I treat human rights as a cross-cutting baseline across all four priorities. Rights should shape safety thresholds, public-sector procurement, labour transition, redress, and oversight. Open-source models and wider social implications are also important, but the first Dialogue will be strongest if it anchors those debates in practical mechanisms that countries can actually use.

Yes. Three cross-cutting issues deserve more explicit attention. First, lifecycle governance for changing systems. Many risks arise after deployment, when systems are fine-tuned, connected to tools, updated, or embedded in new workflows. Governance should therefore address monitoring, change classification, incident reporting, pause or rollback arrangements, and notification for higher-impact changes. UNESCO already emphasizes monitoring across the AI lifecycle, and OECD work points to the need for better incident evidence and common reporting. Second, compute, energy, and infrastructure concentration. AI capacity is shaped not only by talent and data, but by access to chips, cloud, power, cooling, and connectivity. The World Bank and OECD both show that compute supply is concentrated and that many developing countries face high cost, weak infrastructure, and dependency risks. UNEP and UNESCO also highlight AI's environmental footprint across its lifecycle. Third, representational fit: local language coverage, gender, disability, informal work, and cultural context. ITU data show the gender digital divide has stalled globally, and ILO data show women face higher exposure to automation from generative AI. Without local data, evaluation, and meaningful participation, systems may scale exclusion rather than opportunity. The consultation materials also rightly raise the challenge of accountability in an era of AI agents.

From my perspective, especially across the U.S.-India technology and public-interest ecosystem, the biggest governance gap is not a lack of principles; it is the lack of practical, interoperable implementation. In real terms, this affects sectors such as healthcare, finance, education, public services, and software-intensive industries in three ways. First, there is a trust gap. Organizations can deploy AI faster than they can explain it, test it, or correct it after deployment. That weakens public confidence, especially where decisions affect safety, rights, jobs, or access to services. The consultation materials also show this clearly in areas such as law enforcement, civil society participation, and gender inclusion. Second, there is a capacity gap. AI opportunity is growing, but infrastructure, compute, skills, connectivity, and local-language resources remain unevenly distributed. The World Bank notes that many developing countries still face steep barriers in the foundations needed to adapt and deploy AI at scale, while ITU reports that 2.2 billion people remain offline and digital divides persist. Third, there is a workforce and accountability gap. Generative AI can raise productivity, but it also creates uncertainty for clerical and some professional work, making reskilling and human oversight essential. The opportunity is equally significant: if governance becomes clearer, lighter, and more interoperable, countries can expand trustworthy adoption, support local innovation, reduce compliance friction, and ensure AI serves development rather than deepening existing divides. That is where the Global Dialogue can add real value.

The AI Dialogue can play four practical roles. First, it can be the universal convening space where all countries, not only the most technologically advanced, can shape the global conversation. Second, it can be a translation layer: turning a fragmented landscape of legal, technical, human-rights, standards, and development discussions into a shared vocabulary and a clearer picture of where approaches already align. Third, it can be an implementation bridge: connecting the Independent International Scientific Panel's evidence with practical cooperation tools such as capacity partnerships, incident-reporting approaches, procurement guidance, and templates for human oversight and accountability. Fourth, it can be a continuity mechanism that keeps momentum between annual meetings through light thematic tracks, regional inputs, and public progress updates. Its distinctive value is that it is not another negotiating forum and not another expert-only summit. The UN has framed it as an inclusive, non-negotiating platform meant to complement existing efforts and improve coherence across them. That gives it room to do something many forums cannot: bring governments, industry, civil society, academia, and the technical community together without requiring prior legal convergence. In practice, the Dialogue should become the world's coordination layer for AI governance: map convergence, identify gaps, reduce duplication, and mobilize help where states need it most. That would make international cooperation more cumulative, more inclusive, and more usable in the real world.

The Dialogue should build upon and connect with a small number of serious existing pillars rather than start from scratch. These include UNESCO's Recommendation on the Ethics of AI, which provides a global normative baseline on human rights, transparency, accountability, and human oversight; the OECD AI Principles, OECD.AI, and the integrated GPAI partnership, which offer practical policy tools, national policy inventories, expert networks, and incident-related learning; ITU's AI for Good and AI standards work, which connect policy to technical standards, skills, and implementation communities; the Council of Europe Framework Convention on AI, which brings a legally binding human-rights, democracy, and rule-of-law lens; and the Hiroshima AI Process, which has developed risk-management guidance and a code of conduct for advanced AI systems. Regional and thematic processes already linked to the preparatory phase, including Africa-Arab regional dialogue, gender inclusion, civil society, and public trust consultations, should also feed directly into the Dialogue. The added value of the AI Dialogue is connection, legitimacy, and balance. It can provide a UN-backed crosswalk showing where frameworks converge, where they differ, and which elements are most usable for countries with limited capacity. It can also connect political commitments to operational tools, for example by pairing principles with standards, procurement approaches, incident reporting, and capacity support. Most importantly, because every UN Member State has a seat, the Dialogue can bring broader legitimacy and stronger Global South participation than smaller club formats alone. In that sense, it can be the place where scattered AI governance efforts become more coherent, inclusive, and actionable.

Different stakeholders should contribute in distinct but connected roles. Member States should frame public-interest priorities and identify where international cooperation is needed. Industry should bring deployment evidence, safety practices, and implementation lessons. Civil society should surface lived impacts, rights concerns, and accountability gaps. Academia and the technical community should translate research into usable policy options. International organizations should help connect standards, development support, and capacity-building. This is already consistent with the Dialogue's proposed multistakeholder design and consultation process. I would retain the current two-day structure, because it already balances plenary, thematic breakouts, the Scientific Panel, and a "Dialogue of Dialogues." My recommendation is to make it more operational in four ways. First, require each thematic session to end with three outputs: key convergence points, open questions, and 2–3 practical cooperation proposals. Second, appoint a neutral rapporteur for every breakout so stakeholder inputs are captured clearly for the Co-Chairs' summary. Third, combine in-person and remote participation with equal speaking rules, multilingual access, and advance written submissions in all UN languages. Fourth, reserve a small number of speaking slots for evidence from practice: one public-sector case, one civil-society case, one technical case, and one developing-country case per cluster. In short, the Dialogue should not only hear stakeholders; it should assign each of them a function in problem-solving. That would make participation more meaningful and the final summary more actionable.

The most underrepresented voices are often those most affected and least resourced: stakeholders from Least Developed Countries, Landlocked Developing Countries, and Small Island Developing States; civil society from the Global South; local-language communities; women and gender-diverse people; persons with disabilities; workers in informal economies; youth; Indigenous communities; and frontline practitioners in education, health, labour, and local government. The preparatory materials themselves recognize this problem by emphasizing developing-country participation, Global South engagement, gender inclusion, and civil-society input. Inclusion should therefore be designed as authorship, not attendance. That means travel support for underrepresented participants, hybrid participation for those who cannot travel, interpretation and written inputs in all UN languages, accessible formats for persons with disabilities, and structured pathways from regional consultations into the Geneva agenda. It also means dedicating formal speaking slots to community-based evidence, not only high-level institutional statements. The RightsCon and gender consultation notes are especially useful here: they point to bottom-up participation, grassroots engagement, and the need to ensure that women and girls are leaders in shaping governance, not merely subjects of it. A simple test could guide the Dialogue: if a group bears risk, it should have voice; if it has voice, it should also shape outputs. That would make inclusivity real rather than symbolic.

The most effective formats will be those that move from speech-making to structured problem-solving. I would suggest five. First, evidence-to-action sessions: the Scientific Panel presents one finding, then governments, civil society, and technical experts respond with one practical implication each. Second, policy design labs in small groups, where participants co-develop short proposals on topics such as incident reporting, capacity partnerships, or human oversight. Third, implementation clinics, where one country or organization presents a real governance challenge and others offer practical responses. Fourth, regional synthesis rounds, where Africa, Asia, Latin America, the Arab region, Europe, and small-state groupings briefly surface one priority and one cooperation ask. Fifth, commitment boards, where stakeholders can publicly register concrete follow-up actions before 2027. These ideas build directly on the consultation materials' call for innovative formats, practical outputs, and continuity mechanisms beyond the inaugural Dialogue. The current draft structure already contains a strong foundation: thematic breakouts, a Scientific Panel session, side events, and a "Dialogue of Dialogues." I would refine it by making every major session produce a short written output and by allowing brief, well-curated "lightning inputs" from underrepresented stakeholders. The key principle is simple: people engage most seriously when they are asked not only what they think, but what they would do next. That is the format most likely to generate trust, energy, and usable results.

Promising examples already exist, especially when used as a stack rather than as competing models. UNESCO's Recommendation on the Ethics of AI provides a widely shared normative baseline, and UNESCO's Readiness Assessment Methodology (RAM) and Global AI Ethics Observatory help countries assess institutions, laws, skills, and infrastructure before rushing into regulation or deployment. This is a practical model for countries that need a structured starting point rather than abstract principles alone. At the organizational level, NIST's AI Risk Management Framework and its Generative AI Profile are strong examples of lifecycle governance: they help organizations map risks, measure them, manage them, and govern them continuously, including for generative AI. ISO/IEC 42001 adds an auditable management-system approach, which is useful because many governance failures arise from weak processes, not only weak models. For testing and implementation, Singapore's AI Verify is especially valuable because it turns governance principles into a testing framework and software toolkit that organizations can actually use. For cross-border learning, the OECD AI Principles, OECD.AI Policy Observatory, and AI Incidents Monitor are good practice because they support interoperability, policy comparison, and shared learning from failures and near misses. Finally, the Council of Europe Framework Convention on AI and the EU AI Act show how risk-based legal governance can be tied to human rights, safety, and accountability. The main lesson is simple: effective AI governance usually requires five layers working together-principles, readiness assessment, organizational controls, testing, and incident learning. The AI Dialogue could add real value by helping countries adapt these building blocks in low-cost, interoperable ways rather than starting from scratch each time.