Kyiv aviation institute

A successful Global Dialogue on AI Governance should be assessed not by the number of declarations adopted, but by the extent to which it produces operational outcomes. First, it should move beyond abstract principles toward implementable mechanisms such as risk assessment models, liability frameworks, and minimum transparency standards. Without this, discussions risk remaining purely normative. Second, greater alignment between regulatory approaches is essential. The current landscape is fragmented, with different jurisdictions developing parallel frameworks. Even partial convergence (at the level of concepts, definitions, or baseline standards) would reduce regulatory inconsistency and improve legal certainty. Third, the dialogue must ensure substantive multistakeholder participation. This requires that technical experts, academia, and the private sector are not merely present, but actively shape outcomes. Otherwise, the process risks becoming predominantly intergovernmental and less responsive to technological realities. Fourth, success depends on institutional continuity. A one-off event has limited impact unless it leads to sustained formats such as working groups, research collaborations, or permanent coordination platforms. Fifth, the dialogue should meaningfully include underrepresented regions, particularly those not traditionally shaping global regulatory agendas. This is critical for both legitimacy and the long-term viability of governance models. Finally, a key indicator of success is whether the outcomes are transformed into practice through legislation, corporate policies, and international standards. Without implementation pathways, even well-developed frameworks remain symbolic. In this sense, success lies in producing not only shared understanding, but also a clear articulation of next steps, responsibilities, and mechanisms for follow-up.

• AI capacity-building
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Protection and promotion of human rights

My selection reflects a focus on how AI governance can function across jurisdictions while remaining implementable in practice. Interoperability of governance approaches is critical given the current fragmentation of regulatory models. Without at least partial alignment (conceptual, procedural, technical) AI governance risks becoming inconsistent and difficult to operationalize, especially for cross-border systems. Transparency, accountability, and human oversight are foundational for any meaningful governance framework. In my work, these are not abstract values but legal and institutional questions - who is responsible, under what conditions, and how oversight is exercised in complex socio-technical systems. Protection and promotion of human rights remains a baseline that should structure all AI-related regulation. This includes not only classical rights (like privacy or non-discrimination), but also their reinterpretation in digital environments where automated decision-making reshapes power dynamics. Finally, AI capacity-building is essential to avoid asymmetries in participation. From an academic and institutional perspective, there is a clear gap between those who develop regulatory models and those expected to implement or comply with them. Strengthening expertise, particularly in countries with emerging digital ecosystems, is necessary for inclusive and legitimate governance.

One key issue not fully captured is the interaction between AI governance and existing domains such as intellectual property, cybersecurity, and digital infrastructure governance (including DNS and internet governance more broadly). These intersections increasingly determine how AI systems are developed, deployed, and controlled. Another underexplored dimension is the governance of data provenance and training datasets, including questions of legality, ownership, and cross-border data flows. This has direct implications for both accountability and fairness. Additionally, institutional capacity and enforcement mechanisms deserve more attention. Many frameworks focus on principles, but less on how regulatory bodies, courts, or oversight institutions will actually implement and enforce them. Finally, there is a growing need to address AI in contexts of hybrid threats and information security, where systems can be used for manipulation, disinformation, or destabilization. This is particularly relevant for countries operating under conditions of heightened security risk. These issues cut across the listed themes but require more explicit articulation to ensure that governance frameworks remain grounded in real-world institutional and geopolitical contexts.

In Ukraine, governance gaps in AI are shaped by a combination of rapid digitalisation, limited institutional capacity, and the ongoing security context. One of the most significant challenges is regulatory fragmentation and uncertainty. While there is increasing alignment with EU approaches national frameworks remain underdeveloped. This creates ambiguity for both public institutions and private actors regarding compliance, liability, and standards of responsible AI use. A second challenge relates to capacity constraints. There is a clear shortage of interdisciplinary expertise at the intersection of law, technology, and policy. This affects not only regulatory design but also implementation and oversight. Universities and research institutions are beginning to address this gap, but systemic capacity-building is still needed. Transparency and accountability mechanisms are also insufficiently developed. In practice, there are limited tools for auditing AI systems, assessing risks, or ensuring meaningful human oversight, particularly in public sector use. At the same time, the Ukrainian context presents specific security-related challenges, including the use of AI in information operations, cybersecurity threats, and critical infrastructure protection. These issues require integrating AI governance with broader national security and resilience frameworks. Despite these gaps, there are notable opportunities. Ukraine has a highly adaptive digital ecosystem, strong IT sector, and experience in deploying digital public services. This creates a favorable environment for piloting governance models and regulatory sandboxes. Additionally, alignment with EU standards provides a strategic pathway for developing coherent regulation and integrating into broader digital markets. So the current moment shows a dual nature - significant regulatory and institutional gaps, but also a unique opportunity to design AI governance frameworks that are both flexible and resilient in complex conditions.

The AI

There are already strong enough initiatives that are worth building on: OECD AI Principles, UNESCO recommendations, the Council of Europe approach, as well as the EU AI Act, and it is also worth mentioning platforms such as the IGF, where there is already experience in multi-stakeholder dialogue. The problem is that they exist in parallel, and do not always "talk" to each other. Here, AI Dialogue can be useful as a place where these approaches are brought together to make it clear where they coincide and where they do not. Another important point: the inclusion of those who are usually not heard much. Currently, rules often form several centers of power, but then everyone must apply them. It is also important to talk not only about principles, but about how it works in practice: cases, experience, mistakes. And separately - the connection with other areas: cybersecurity, data, intellectual property. Without this, AI governance looks disconnected from reality. Therefore, the added value of such a dialogue lies not in new declarations, but in gathering existing approaches into a more holistic and understandable picture.

Different stakeholders should not merely be present; they should also fulfill clearly defined roles. States set the political and regulatory vector, but should not monopolize the discussion. Business provides practical experience in the development and implementation of AI systems. The technical community helps to understand what is actually possible from a technological point of view and what is not. The academia provides analytical depth and a critical assessment of long-term consequences. The format should be focused not on speeches, but on work. Optimally, it will look like this: -short plenary sessions to pose problems; - thematic working groups to discuss specific issues (liability, risk assessment, data, etc.); - final sessions with clear conclusions and next steps. An important element is the continuation after the event: working groups, joint developments, short analytical documents, etc. Effective participation occurs when stakeholders do not just voice positions, but work together to solve specific problems.

Global discussions on AI are still largely shaped by a limited number of actors (primarily technologically advanced countries and large companies). Currently underrepresented are: - countries with transitional or developing digital ecosystems; - academic communities outside leading global centers - public sector representatives who directly implement policies; - local human rights and civil society organizations. The problem is not only in access, but also in the ability to fully participate. To improve the situation, it would be advisable to: - provide prior preparation of participants (materials, briefings, training, etc.); - guarantee balanced representation in panels and working groups; - create formats where smaller actors can speak out, not just listen; - hold regional preliminary discussions to gather positions. It is also important to consider institutional diversity: practitioners from the "field" often see problems that are not visible at the global level. Without these voices, AI governance risks becoming disconnected from the real-world context of application.

For dialogue to be effective, it is necessary to move away from the passive listening format to active collaboration. One of the most effective approaches is case-oriented discussions. Instead of general principles, participants work with specific situations (for example, the use of AI in public services or information security). Here, academia can play a key role - to form such cases, structure them and set a framework for analysis. The second format is working laboratories, where small groups prepare short results: recommendations, risk assessment models or regulatory approaches, and the like. The academic community can act as a moderator of such processes and ensure their methodological quality. It is also important to mix participants from different fields in small groups. In this context, academia can act as a "translator" between sectors - to explain complex technical or legal aspects and reduce different positions to a common understanding. It is advisable to use digital tools: - real-time collection of positions and priorities; - joint editing of documents; - short feedback cycles after sessions, etc. In addition, it is worth limiting long speeches and leaving more time for discussion. Separately, the academia can ensure further understanding of the results (preparation of analytical materials, policy briefs and studies that will give the dialogue a continuation). The general logic is simple: less declarativeness - more joint work on specific solutions with a clear analytical basis.

There are already several approaches that show how AI governance can work in practice. The EU AI Act is probably the clearest example. Its strength is a risk-based approach; not all AI is regulated the same way, only systems that can actually create high risks. This makes regulation more adequate and applicable. The OECD AI Principles are less about hard law, but important as a common basis. Many countries are guided by them, and this helps to at least partially speak "the same language". From a practical point of view, such tools as algorithmic impact assessments (for example, in Canada) are interesting. This is when, before implementing AI, they assess what risks it can create. This is a simple but very logical mechanism. The topic of AI auditing is also actively developing - checking systems for compliance with requirements. This is important, because there is always a gap between "we wrote the rules" and "they really work". If we talk about Ukraine, there is strong potential here. We have a very flexible IT environment and a strong academic base that is starting to actively engage in these discussions. Our experience working in risk environments is especially valuable - cybersecurity, disinformation, infrastructure protection. This is not theory, but daily practice. And this experience can be useful at the global level. Therefore, effective AI governance is not only about large-scale regulations, but about a combination of law, practice and normal academic analytics that helps to connect it all.