COPA

A successful first Global Dialogue would move AI governance from broad principles toward practical cooperation that institutions can actually implement. From my perspective as founder of COPA, success would mean three outcomes. First, the Dialogue should establish a stronger shared baseline for responsible AI deployment across sectors. This should include safety, accountability, transparency, meaningful human oversight, and clear expectations for how AI systems are evaluated in real organizational settings. Second, it should create an implementation pathway, not only a normative statement. Many governments, international development organizations, multilateral institutions, and private sector actors want to use AI responsibly, but need practical tools for procurement, risk assessment, monitoring, capacity-building, and institutional governance. The Dialogue should help translate global principles into usable guidance. Third, inclusion should be operational, not symbolic. Smaller states, under-resourced institutions, non-dominant language communities, and organizations working in complex institutional contexts should shape the agenda. Their needs around access, local relevance, linguistic inclusion, and institutional readiness are central to whether global AI governance will work in practice. I would judge the Dialogue's success not by the quality of speeches alone, but by whether it creates durable channels for cooperation, technical exchange, and policy follow-through. If participants leave with concrete areas for coordination, a stronger shared vocabulary, and a credible roadmap for continued work, the first Dialogue will have achieved meaningful impact.

• Safe, secure and trustworthy AI
• AI capacity-building
• Interoperability of governance approaches
• Open-source software, open data and open AI models

I selected these priorities because they are central to making AI governance practical, inclusive, and implementable across different institutional contexts. Safe, secure and trustworthy AI is foundational. At COPA, our work focuses on turning complex organizational challenges into structured, evidence-based solutions. That is not possible without confidence that AI systems are reliable, appropriately used, and governed with clear safeguards. AI capacity-building is equally urgent. Many governments, development organizations, multilateral actors, and private sector institutions are interested in AI, but do not yet have the internal capacity to evaluate systems, design governance processes, manage risks, or use AI responsibly. Without capacity-building, global AI governance will remain uneven. Interoperability of governance approaches matters because AI systems, vendors, data flows, and institutional partnerships often cross borders. Fragmented rules and terminology make responsible adoption harder, especially for smaller markets and organizations operating internationally. Greater alignment can improve cooperation while still allowing for local adaptation. Open-source software, open data, and open AI models are also important because they can broaden access to innovation, support transparency, enable local adaptation, and reduce dependence on a small number of dominant actors. However, openness should be paired with responsible use practices, security considerations, and clear governance safeguards. Together, these priorities support a model of AI governance that is trustworthy, operational, collaborative, and accessible to institutions with different levels of capacity.

Yes. One important cross-cutting issue is the concentration of power across the AI stack, including compute, cloud infrastructure, foundation models, and access to high-quality data. Governance discussions often focus on the use of AI systems, but the distribution of underlying infrastructure strongly shapes who can innovate, compete, and meaningfully participate in governance. A second issue is the gap between frontier AI governance and deployment governance. Many institutions are not building frontier models, but they are rapidly adopting AI tools for research, analysis, decision support, public communication, service delivery, and organizational improvement. Their governance needs are different. They require practical support on procurement, local evaluation, monitoring, human oversight, accountability, and change management. A third issue is multilingual and low-resource context performance. AI systems may perform well in dominant languages while underperforming in smaller linguistic or regional contexts. This has direct implications for safety, fairness, accessibility, and public trust. Another emerging issue is institutional dependency. Governments and organizations may become reliant on external AI providers without sufficient ability to evaluate, adapt, or govern those systems locally. This creates long-term strategic, operational, and accountability risks. From COPA's perspective as an applied AI consultancy, these issues matter because responsible AI governance must work in real institutions, not only at the level of global principles. They cut across safety, openness, interoperability, and capacity-building, and they will determine whether AI governance becomes genuinely inclusive and practical.

In my region and sector, the main challenge is that interest in AI adoption is advancing faster than institutional capacity to govern it. Across public institutions, international development organizations, multilateral partners, and private sector actors, there is growing demand to use AI for analytics, decision support, public communication, research, and operational improvement. However, many organizations still lack clear internal standards for safety, evaluation, procurement, accountability, and oversight. This creates two risks. Some institutions may adopt AI tools without sufficient governance, while others may avoid useful innovation because they do not trust their ability to assess and manage the risks. In both cases, the governance gap limits the ability of AI to deliver measurable and responsible impact. Capacity-building is therefore central. Technical talent may exist, but institutional readiness is often uneven. Organizations need support not only with tools, but with processes, policies, operating models, and evidence-based decision frameworks. Interoperability is also important for smaller markets and cross-border work. When governance approaches diverge sharply across jurisdictions, local actors face uncertainty and higher compliance burdens, especially when they rely on external models, platforms, or partnerships. Openness presents both an opportunity and a challenge. Open-source software, open data, and open models can support access, transparency, and local adaptation, including in multilingual and lower-resource contexts. But many institutions need practical guidance to use open tools responsibly and securely. Overall, these governance gaps affect risk management, competitiveness, inclusion, and the ability of institutions to benefit from AI in ways that are locally relevant and sustainable.

The AI Dialogue can play an important role by creating a coherent space for international cooperation across currently fragmented efforts. Many actors are already working on AI governance through regulatory, technical, regional, sectoral, and institutional processes. The Dialogue can help connect these efforts without replacing them. Its most important contribution would be to support convergence where convergence is useful. This includes common terminology, shared safety principles, better coordination on capacity-building, and more interoperable governance approaches across jurisdictions. Cooperation does not require full uniformity, but it does require enough shared understanding to make coordination possible. The Dialogue can also help balance frontier AI concerns with practical deployment needs. Many governments, development organizations, and private sector institutions are not building the largest models. They are trying to use AI responsibly in complex organizational settings. Their experience should inform global cooperation, especially on procurement, evaluation, oversight, accountability, and implementation capacity. Another key role is to create a constructive space for openness. Open-source software, open data, and open AI models are central to access and innovation, but the debate can become polarized between unrestricted openness and overly restrictive approaches. The Dialogue can help identify balanced frameworks that support access while managing risk. From COPA's perspective, the Dialogue would add real value if it reduces duplication, improves mutual learning, and turns high-level principles into guidance that institutions can use. International AI governance should be practical enough to support implementation, not only agreement in principle.

The AI Dialogue should build on existing work rather than start from zero. Useful reference points include UNESCO's Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the Global Digital Compact, technical standard-setting efforts, and regional regulatory developments such as the European Union's AI governance frameworks. It should also connect with practitioner communities working on AI assurance, model evaluation, public-sector digital transformation, and responsible deployment. The added value of the Dialogue would be different from that of any single existing initiative. Its comparative advantage is convening power across geographies, institutions, and sectors. It can connect normative discussions with operational realities and create a space where governments, international organizations, researchers, companies, civil society, and implementation practitioners identify concrete areas of alignment. It can also elevate implementation challenges that are often underemphasized. These include AI capacity-building, multilingual and low-resource contexts, responsible public-sector procurement, organizational readiness, and the governance need of institutions that deploy AI systems rather than build frontier models. A well-designed Dialogue should not duplicate standards or regulation. Instead, it should help map where existing efforts complement each other, where gaps remain, and where coordination could reduce fragmentation. Its contribution should be to improve coherence, increase inclusion, and accelerate the practical usability of global AI governance efforts for institutions working in complex real-world environments.

Different stakeholders should contribute in distinct but complementary ways. Governments can bring public-interest priorities, regulatory experience, and national implementation challenges. International development and multilateral organizations can contribute experience from complex institutional settings and capacity-building work. Companies can share deployment realities, risk-management practices, and technical lessons from building or using AI systems. Researchers can provide evidence, evaluation methods, and independent analysis. Civil society can surface human rights, labor, inclusion, and accountability concerns that might otherwise be overlooked. To make these contributions meaningful, the Dialogue should balance plenary visibility with smaller, output-oriented working formats. High-level sessions are useful for political signaling, but most substantive progress will come from moderated thematic tracks, practical roundtables, and breakout groups with clear questions and deliverables. The process should also include pre-dialogue consultations and written submissions so participation is not limited to those physically present. Regional and sector-specific consultations in advance would improve the quality of discussion and broaden representation. I would recommend a three-layer format: a plenary level for shared direction, thematic working sessions for substance, and a follow-up mechanism for continuity. Each thematic session should produce a short synthesis of key gaps, points of convergence, and next steps. This would help ensure that the Dialogue is not only broad and inclusive, but cumulative and useful for future cooperation.

Several perspectives remain underrepresented in global AI governance discussions, particularly those of smaller states and non-dominant language communities. As an Armenian practitioner working in policy and AI implementation, I see a clear gap in how the needs of countries like Armenia are reflected in these debates. AI governance is often shaped by large markets, major technology producers, and institutions with significant regulatory capacity. Smaller states, however, face different and very practical questions: how to build institutional readiness, procure AI systems responsibly, protect public data, evaluate tools in local languages, and ensure that AI serves public-sector needs without creating new forms of dependency. Non-dominant language communities are also insufficiently represented. For Armenian and other lower-resource languages, AI governance cannot be separated from questions of language access, model performance, cultural context, and the risk of exclusion from AI-enabled services. If systems do not work reliably in local languages, citizens will not benefit equally from them. Another missing perspective is that of practitioners responsible for implementation: public administrators, educators, health-system managers, SME leaders, analysts, and operational teams. These are the people who encounter the real constraints of adoption, including skills gaps, trust, accountability, cybersecurity, and workflow integration. To include these voices, participation must be intentional. This means funding support for attendance, multilingual formats, regional consultations, genuinely interactive remote participation, and selection criteria that value implementation experience, not only institutional prestige. It would also help to reserve speaking and working-group roles for smaller states, local-language communities, and practitioners from public-sector and civil-society contexts. Inclusion should be treated as a design requirement of AI governance, not as a symbolic gesture.

To foster meaningful engagement, the Dialogue should combine formal discussion with formats that are participatory, practical, and inclusive. One useful approach would be to organize small, mixed-sector working sessions around key AI governance challenges. These sessions could bring together governments, civil society, technical experts, private-sector actors, and implementation practitioners to discuss issues such as public-sector use of AI, accountability, language access, institutional capacity, and risk management. The aim should be to move beyond general statements and create space for more grounded exchange. Another effective format would be structured case discussions. Participants could reflect on real or realistic examples of AI deployment, oversight, or policy design, focusing on what worked, what failed, and what lessons can be applied across different contexts. This would help connect high-level governance principles with practical experience. The Dialogue could also include regional and thematic sessions that allow different communities to articulate their priorities before broader plenary discussions. This would make participation more balanced and help ensure that smaller states, non-dominant language communities, and implementation-focused actors are not only present but meaningfully heard. Interactive formats should also be supported by strong preparation and follow-up. Pre-circulated questions, short written inputs, moderated working sessions, and post-event thematic groups could help participants engage more substantively. Remote participation should also be designed to allow active contribution rather than passive observation. Overall, the most effective engagement formats would be those that make the Dialogue less one-directional and more collaborative. Meaningful engagement requires not only speaking opportunities, but structured ways for diverse participants to shape the agenda, exchange practical experience, and contribute to outcomes that continue beyond the event itself.

AI governance or offer concrete solutions to addressing its challenges. (Max. 300 words) Effective AI governance depends on practical systems that help institutions use AI responsibly in everyday work. A useful starting point is risk-based governance. Different AI tools create different levels of risk, so the rules, checks, and oversight should reflect the use case, the people affected, and the possible impact. Governance should also cover the full lifecycle of an AI system, including design, procurement, testing, deployment, monitoring, and post-deployment review. This requires clear documentation, evaluation processes, escalation channels, and ownership. Human oversight is especially important in sensitive areas. Institutions need clear roles, review points, intervention thresholds, and escalation procedures so that accountability is built into the process. Regular evaluation is another key practice. AI systems should be assessed for accuracy, bias, security, ability to be explained, and performance in the relevant language and institutional context. Capacity building is also essential. Decision-makers, analysts, and operational teams need practical training on both the opportunities and limits of AI. From COPA's perspective, the strongest approaches are those that turn broad governance principles into everyday institutional practice and help organizations make better, evidence-based decisions.