Ukrainian-French Institute for Science, Innovation & Economic Development

The first Global Dialogue on AI Governance will be successful only if it moves beyond broad consensus and produces a credible path from principles to implementation. First, it should affirm a clear normative baseline: AI governance must be grounded in human rights, human dignity, transparency, accountability, and meaningful human oversight. Without this, "governance" risks becoming either a technical exercise or a competition of national interests. Second, the Dialogue should identify a small number of actionable priorities that states and institutions can actually implement, including those with limited regulatory and technical capacity. Success would mean not only agreeing on values, but also translating them into practical safeguards, institutional responsibilities, and implementation tools. Third, it should correct a major weakness in current global discussions: the underrepresentation of people and contexts where governance failures are most damaging. Displaced populations, conflict-affected communities, low-capacity institutions, and linguistically marginalized groups must not remain peripheral to AI governance debates. Their realities should shape the agenda itself. Fourth, the Dialogue should strengthen interoperability across governance approaches. The goal is not uniformity, but sufficient coherence to avoid rights gaps, fragmented safeguards, and unequal levels of protection. Finally, success would mean creating a follow-up architecture: a roadmap, channels for inclusive participation, and measurable outputs that allow the Dialogue to evolve into a sustained governance process rather than a one-time event. In short, success would mean building governance that is not only globally discussed, but institutionally usable, socially legitimate, and responsive to real human vulnerability.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Interoperability of governance approaches

I selected these four priorities because they address the core question of whether AI governance will protect people in practice, not only regulate systems in principle. Protection and promotion of human rights must remain central because AI increasingly shapes access to services, information, mobility, education, work, and identity. If governance does not begin from rights, it can easily normalize exclusion, discrimination, and unequal protection. Transparency, accountability, and human oversight are essential because affected individuals and institutions must be able to understand decisions, challenge harmful outcomes, and seek remedy. This is particularly important where AI systems are used in public services or contexts marked by vulnerability and power imbalance. I also selected the social, economic, ethical, cultural, linguistic and technical implications of AI because AI is never neutral in its effects. It interacts with existing inequalities, language hierarchies, institutional weaknesses, and uneven digital access. Governance must therefore reflect lived realities, not only system performance. Finally, interoperability of governance approaches is critical because fragmented rules create uneven safeguards and leave less-resourced countries and communities exposed. Global AI governance does not require identical models, but it does require enough coherence to make protections portable, credible, and internationally relevant. Together, these four priorities help ensure that AI governance remains human-centered, operational, and inclusive across very different national and social contexts.

Yes. One critical cross-cutting issue is AI governance in contexts of displacement, conflict, and post-crisis recovery. Much of the global discussion still assumes stable institutions, reliable documentation, high digital literacy, and relatively equal access to services. Yet many of the people most affected by digital systems live in conditions of fragility: refugees, displaced persons, stateless populations, and communities affected by war or institutional breakdown. In such settings, AI-related harms are often intensified through exclusion, inaccessible systems, identity mismatches, language barriers, weak feedback channels, and limited access to remedy. A second issue is the need for inclusion-by-design metrics. Governance discussions often focus on safety, innovation, or compliance in broad terms, but do not sufficiently measure whether systems actually reduce barriers for vulnerable users. It is not enough to ask whether a system functions; we must ask for whom it functions, under what conditions, and at whose cost. A third issue is institutional readiness. Many public institutions are expected to govern AI without the expertise, resources, or implementation tools necessary to do so effectively. Capacity-building should therefore be linked to institutional capability, local context, and public-interest implementation. In my view, the credibility of global AI governance will be tested precisely in these contexts of fragility. If governance frameworks cannot protect those facing the highest barriers, they will remain normatively persuasive but operationally incomplete.

In my sector, the most significant AI governance gaps are appearing where vulnerability is highest: in public services, migration-related systems, post-conflict recovery, and institutions operating under pressure. The first challenge is uneven protection. While AI adoption is accelerating, safeguards are not developing at the same speed. In practice, this means that affected individuals often face opaque decisions, weak accountability, inaccessible complaint mechanisms, and limited human review. These risks are particularly serious for displaced persons, refugees, linguistically marginalized groups, and people with interrupted documentation or low digital literacy. The second challenge is institutional asymmetry. Some governments and organizations are advancing AI strategies, but many public institutions still lack the capacity to assess risks, implement oversight, or align innovation with human rights obligations. This creates a gap between policy ambition and operational reality. The third challenge is fragmentation. Different governance models are emerging across jurisdictions, but without sufficient interoperability, rights protections may become uneven and difficult to apply across borders. This is especially relevant in Europe and in contexts linked to migration, mobility, and digital identity. At the same time, there are important opportunities. AI governance can help build more inclusive public systems if it is designed around transparency, accountability, human oversight, and real-world accessibility. There is also an opportunity to embed inclusion metrics into governance frameworks, so that success is measured not only by technical efficiency, but by whether systems reduce barriers for those most at risk of exclusion. In my view, the central test is this: whether AI governance will merely accompany innovation, or whether it will actively shape it in ways that protect dignity, strengthen institutions, and reduce inequality.

The AI Dialogue can play a unique role by creating a trusted multilateral space where states, International Organisations, academia, civil society, and industry can move from parallel debates toward greater coherence in AI governance. The UN's current AI architecture already points in this direction, including the High-level Advisory Body's call for more inclusive international governance and the Global Digital Compact's push for shared global approaches. Its first value is normative: the Dialogue can help reinforce a common baseline around human rights, human dignity, transparency, accountability, and meaningful human oversight, while building on existing global standards rather than reopening first principles each time. UNESCO's Recommendation on the Ethics of AI and the OECD AI Principles already provide important foundations for this work. Its second value is practical: the Dialogue can help bridge fragmented governance approaches by identifying areas where interoperability is both feasible and necessary, especially for public-interest use cases, cross-border risks, and low-capacity contexts. The aim should not be uniformity, but enough alignment to reduce regulatory gaps and unequal protection. Its third value is political and institutional: the Dialogue can ensure that countries and communities often underrepresented in AI governance discussions have a meaningful role in shaping implementation priorities. This is particularly important if global cooperation is to remain credible beyond the most resourced jurisdictions. In short, the Dialogue can advance international cooperation by serving as a connector: between principles and practice, between existing frameworks, and between global norms and local realities.

The AI Dialogue should build on existing initiatives that already provide substantive normative, policy, and institutional foundations. First, it should connect with the UNESCO Recommendation on the Ethics of Artificial Intelligence, which remains the most universal intergovernmental standard on AI ethics, adopted by all UNESCO Member States. Second, it should build on the OECD AI Principles, which have become an important reference point for trustworthy AI and international policy coordination. Third, it should align with the UN Secretary-General's AI Advisory Body and the follow-up architecture emerging from the Global Digital Compact, including the broader UN effort to strengthen global cooperation on AI governance. It should also remain connected to emerging work on interoperability in AI safety and governance, since one of the central global challenges is not the absence of initiatives, but the fragmentation between them. The added value of the AI Dialogue should therefore not be to duplicate existing frameworks. Its real value would be to act as a coordinating layer: connecting norms with implementation, linking technical and rights-based approaches, and identifying where common safeguards, institutional support, and shared priorities can realistically be advanced. Most importantly, the Dialogue can bring inclusive political legitimacy. It can create a space where underrepresented countries, vulnerable populations, and lower-capacity institutions are not merely consulted, but help shape how global AI governance is operationalized in practice.

Different stakeholders should contribute according to their institutional role, but within a format that values complementarity rather than hierarchy. Member States can bring regulatory priorities, implementation realities, and public accountability. International Organisations can connect AI governance to existing multilateral frameworks and identify areas for coordination. Academia and independent researchers can contribute evidence, conceptual clarity, and critical analysis. Civil society can surface rights-based concerns, especially from communities that experience exclusion first. Industry and technical actors can provide operational insight, but should be engaged as contributors to public-interest governance, not as agenda-setters. To make this effective, the AI Dialogue should be structured in layers. First, it should include a plenary level for shared priorities and political direction. Second, it should have thematic working sessions designed around concrete governance problems rather than abstract themes for example, accountability in public services, AI in fragile contexts, or barriers to remedy. Third, it should include structured stakeholder roundtables where each group is invited to respond to the same governance question from its own vantage point. Fourth, it should produce short synthesis outputs after each session: areas of convergence, open disagreements, and actionable next steps. The Dialogue should also include mechanisms for written input before and after meetings, so participation is not limited to those physically present or best resourced. In short, the format should be multistakeholder, but also disciplined: inclusive in participation, problem-oriented in structure, and designed to generate usable outcomes rather than symbolic exchange.

The most underrepresented voices in global AI governance are often those who live closest to governance failure. These include refugees, displaced persons, stateless populations, low-income and low-literacy users, people with disabilities, linguistically marginalized communities, educators and local public-service practitioners, and institutions operating in post-conflict or low-capacity settings. Their realities are rarely central, even though they are among the first to experience exclusion, inaccessible systems, weak remedy mechanisms, and the harmful effects of poor design. Another underrepresented group is public-interest expertise from outside the main regulatory and technical centers of power. Too often, global AI governance is shaped by a narrow set of actors from highly resourced jurisdictions, while countries and communities facing the sharpest governance challenges are invited only after priorities have already been defined. Inclusion should therefore be designed, not assumed. This means providing financial support for participation, multilingual access, hybrid and asynchronous engagement formats, and simplified submission channels for organizations that may not have the resources to engage in highly formalized processes. It also means moving beyond symbolic representation. Underrepresented groups should not appear only in "inclusion panels"; they should be present in agenda-setting, drafting, and follow-up mechanisms. Finally, inclusion should be tied to subject matter. If a session concerns digital identity, social protection, education, migration, conflict, or access to remedy, then communities directly affected by those systems should be structurally included in the discussion. A credible global AI dialogue must not only speak about vulnerable populations. It must be shaped with them.

Meaningful engagement is more likely when participants are asked to work through real governance problems, not simply restate general positions. For that reason, the AI Dialogue should use formats that combine deliberation, comparison, and problem-solving. One effective format would be scenario-based policy labs. Participants from different sectors could respond to the same hypothetical case for example, an AI-assisted public service system affecting displaced populations, or a cross-border governance failure involving biometric identification. This would reveal where principles hold, where they diverge, and what safeguards are missing. A second useful format would be "structured response panels," where states, civil society, researchers, and industry each respond in turn to a single governance question. This can prevent fragmented discussion and make points of convergence more visible. A third format would be implementation clinics: smaller sessions where participants examine a concrete governance challenge such as human oversight, grievance mechanisms, or institutional capacity, and identify practical obstacles as well as workable solutions. The Dialogue could also benefit from pre-session written briefs and post-session synthesis notes, allowing deeper participation from stakeholders who cannot intervene live. Digital participation tools should support multilingual input, short submissions, and thematic clustering of recommendations. Finally, one innovative element would be an "inclusion check" for every session: a brief reflection on whose perspective is missing, whose risks are not visible, and whether the proposed solution would still work in low-capacity or high-vulnerability settings. The strongest engagement formats are those that make dialogue more disciplined, more comparative, and more accountable to real-world use.

Effective AI governance is already being advanced through a number of complementary policy and implementation approaches. A strong example is the UNESCO Recommendation on the Ethics of Artificial Intelligence, which provides a global normative framework grounded in human rights, human dignity, transparency, fairness, and human oversight. Its added value is that it is universal in scope and supported by practical implementation tools for Member States. A second important example is the OECD AI Principles, especially after their 2024 update. They remain one of the most operational international reference points for trustworthy AI and are useful because they combine normative guidance with policy usability and cross-country interoperability. A third example is the Council of Europe Framework Convention on AI and human rights, democracy and the rule of law, which is especially significant because it is the first international legally binding treaty in this field. Its value lies in translating broad principles into state obligations connected to transparency, accountability, equality, and oversight. At the regional level, the EU AI Act is also important because it offers a concrete risk-based regulatory model and creates institutional mechanisms for implementation, including the European AI Office and a dedicated information platform. In my view, the most promising approaches are those that combine four elements: a human-rights foundation, risk-based regulation, real accountability mechanisms, and implementation support for lower-capacity institutions. Good AI governance is not only about setting standards; it is about ensuring that those standards can be applied in practice, including in fragile, unequal, and cross-border contexts.