Insight Research Ireland Centre for Data Analytics

A successful first Global Dialogue on AI Governance should be judged by whether it produces a small set of practical outputs, not only a general statement of principles. The Dialogue should issue a short outcome document that clearly identifies the main governance bottleneck: regulators often lack access to the technical information needed to verify compliance with frontier AI systems. Recognizing this information asymmetry would give the process a more realistic foundation. Besides, the Dialogue should agree to create an inter-sessional working group before the 2027 meeting to develop options for funding AI oversight. This work should include beneficiary-pays approaches, such as supervisory fees or risk-based levies on frontier AI providers. That would help ensure that governance costs are not borne only by public budgets, especially in smaller or lower-capacity states that rely on AI systems developed elsewhere. Moreover, the Dialogue should outline a basic set of shared expectations for frontier AI providers. At a minimum, these should include disclosure of evaluation methods, reporting of serious incidents, documentation of major model updates, and secure access for regulators or accredited auditors where appropriate. It should launch a concrete capacity-building track for under-resourced countries. Useful first steps would include a shared roster of technical experts, training for regulators, and exploration of regional or pooled audit support. Finally, success would require a timeline for follow-up, including a progress note before the 2027 session. Without concrete steps on information access, funding, and capacity-building, the Dialogue risks remaining a valuable conversation rather than becoming a credible mechanism for global AI governance.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI

My priorities are transparency, accountability and human oversight; safe, secure and trustworthy AI; interoperability of governance approaches; and the social and economic implications of AI. These areas are closely linked. In my view, the most urgent governance challenge is not the lack of principles, but the difficulty of enforcing them in practice. Frontier AI systems are developed and operated by a small number of firms that retain most of the technical information needed to assess compliance. Without transparency, access to relevant information, and meaningful accountability, safety requirements remain difficult to verify. Safe and trustworthy AI therefore depends on practical oversight tools, including evaluation standards, incident reporting, and independent scrutiny where appropriate. Interoperability is also essential because AI systems and providers operate across borders. Greater alignment between governance approaches can reduce fragmentation, lower compliance uncertainty, and help smaller or lower-capacity jurisdictions participate more effectively. Finally, the social and economic implications of AI are a priority because the benefits of advanced AI are highly concentrated, while the governance and adjustment costs are more widely distributed. Effective AI governance should therefore focus not only on innovation and risk, but also on fairness, capacity-building, and the sustainability of oversight.

Yes. A key cross-cutting issue is the governance capacity gap. Many states, especially smaller or lower-capacity ones, do not have the technical expertise, infrastructure, or funding needed to supervise advanced AI systems effectively. This links directly to a second issue: information asymmetry. Providers of frontier AI systems usually control the evaluations, documentation, model updates, and deployment data that regulators would need in order to verify compliance. A third issue is the concentration of frontier AI capabilities in a small number of firms and countries. This concentration creates an imbalance in both economic power and governance influence. It also raises the question of who should bear the cost of oversight. For that reason, the Dialogue should consider sustainable funding models for AI governance, including beneficiary-pays or risk-based supervisory fees for the most capable systems. Another emerging issue is post-deployment change. AI systems can be updated, fine-tuned, or integrated into new contexts after release, which means governance cannot rely only on one-time assessments. Ongoing monitoring and incident reporting will be increasingly important. In my view, these issues deserve more explicit attention because they determine whether AI governance frameworks can be implemented in practice rather than remaining only high-level commitments.

In my view, these governance gaps are especially significant for countries and sectors that are primarily adopters, rather than developers, of frontier AI. The main challenge is information asymmetry. Advanced systems are often introduced by a small number of firms and countries, while regulators, businesses, and communities in smaller markets have limited visibility into training data, evaluation methods, model updates, and safety practices. This makes transparency, accountability, and human oversight difficult to implement in practice. A second challenge is capacity. Many public institutions and sectoral regulators do not yet have enough technical expertise, auditing infrastructure, or funding to assess safety claims, investigate incidents, or monitor post-deployment changes. Differences across national governance approaches can further increase compliance costs and make it harder for smaller actors to participate on fair terms. The social and economic effects are also uneven. AI can improve productivity, public services, and access to knowledge, but it can also deepen dependence on foreign providers, concentrate value outside the local economy, and widen gaps between large and small firms or between digitally advanced and less prepared communities. At the same time, there is a major opportunity. Because many countries are still shaping their AI governance approaches, they can build more interoperable, risk-based, and accountable systems from the start. International cooperation, shared technical resources, and sustainable funding for oversight could help smaller or lower-capacity jurisdictions participate more effectively and benefit more fairly from AI. For that reason, the most urgent need is not only more principles, but more practical capacity to verify, adapt, and enforce them.

In my view, the AI Dialogue can add the most value by serving as the United Nations coordination layer across a crowded and uneven governance landscape. Resolution 79/325 gives the Dialogue an inclusive mandate to bring together governments and stakeholders to discuss cooperation, share lessons, and facilitate open and transparent discussion, while the same UN process created an Independent International Scientific Panel to provide evidence-based assessments. The Dialogue should use that architecture to bridge three gaps: between scientific evidence and policy decisions, between frontier AI developers and countries that mainly adopt AI systems, and between regional or club-based initiatives that risk drifting apart. To make that role concrete, the Dialogue should focus on a small set of cooperative tasks. It should identify baseline areas for interoperability, including common terminology, incident reporting, model documentation, evaluation disclosure, and post-deployment monitoring. It should support capacity-building for lower-resource regulators through shared expert rosters, training, and regional technical support. It should also open a practical discussion on sustainable funding for oversight, including beneficiary-pays approaches for the most capable systems, so that governance costs do not fall only on public budgets with the least capacity. Existing initiatives already develop norms, standards, and reporting tools. The AI Dialogue's value is to connect them in a more inclusive forum and help turn fragmented principles into workable cooperation.

The Dialogue should build on several initiatives that already provide important pieces of the governance puzzle. At the UN level, the Global Digital Compact and Resolution 79/325 provide the political mandate for both the Dialogue and the new Independent International Scientific Panel. UNESCO's Recommendation on the Ethics of AI offers the broadest global normative baseline. The OECD AI Principles, updated in 2024, and the integrated GPAI-OECD partnership provide practical policy guidance, common concepts, and tools for interoperability. The Council of Europe's Framework Convention offers the first international legally binding AI treaty and is open beyond Europe. The G7 Hiroshima AI Process and its Reporting Framework add useful experience on frontier model reporting, while ITU's AI for Good ecosystem contributes standards, skills, and capacity-building. The added value of the AI Dialogue should not be to duplicate these efforts. Its value is to connect them in a universal forum, identify where their approaches already converge, and surface gaps that affect countries outside OECD, G7, or European processes. It can also channel evidence from the new UN Scientific Panel into a broader policy discussion and keep attention on implementation challenges that many initiatives address only indirectly, especially regulatory capacity, access to technical information, and sustainable funding for oversight. That combination of inclusiveness, coherence, and practical follow-through would be a meaningful UN contribution.

Different stakeholders should contribute in different ways. Governments should identify priority governance gaps and areas where international cooperation is most needed. Frontier AI companies should provide concrete operational input on evaluation practices, incident reporting, model updates, and what information can realistically be shared with regulators. Researchers and the Independent International Scientific Panel should translate technical evidence into policy-relevant findings. Civil society, workers, educators, consumer groups, and affected communities should highlight where governance is not working in practice. Smaller and lower-capacity states should be able to define the capacity-building and interoperability support they need most. The UN process is already designed as an inclusive, multi-stakeholder platform, with consultations across locations and online, and the July 2026 Dialogue is expected to include a plenary, a high-level governmental segment, thematic discussions, and presentation of the Panel's inaugural report. For structure, I would recommend three layers. First, a short plenary should frame the main governance gaps and hear the Scientific Panel's evidence. Second, thematic roundtables should be small, moderated, and balanced across stakeholder groups, with written questions circulated in advance. Third, each thematic session should end with two or three concrete outputs, such as cooperation options, capacity needs, or follow-up tasks. The Dialogue should also include hybrid multilingual participation, transparent speaker selection, published summaries of written submissions, and intersessional working groups that continue work before the 2027 session. That would make participation more meaningful and outcome-oriented.

In my view, the most underrepresented voices are regulators from smaller and lower-capacity states, public-interest technologists from the Global South, workers and trade unions, small and medium-sized enterprises, educators, health and social service providers, disability communities, linguistic minorities, Indigenous communities, and people directly affected by automated decision-making. Global AI debates are often shaped by the countries and firms that develop the most advanced systems, while countries and sectors that mainly adopt these systems have less influence over the rules and fewer resources to evaluate them. This matters because the UN mandate for the Dialogue is not only to discuss AI governance in general, but also to support sustainable development and help close digital divides between and within countries. Inclusion should therefore be designed, not assumed. The Dialogue should reserve speaking slots for underrepresented groups, publish transparent selection criteria for panel participation, and fund travel and remote participation support for lower-resource stakeholders. It should provide interpretation, accessible formats, and regional pre-dialogue consultations so that people can contribute before the main event. It would also help to require balanced panels that include at least one representative from a lower-capacity jurisdiction and one representative from an affected community. Without these measures, multi-stakeholder participation can become formal rather than genuinely representative.

The most effective formats would be those that move beyond prepared statements and create structured problem-solving. One useful format would be short evidence briefings from the Independent International Scientific Panel, followed by responses from a government representative, a company, and a civil society or worker representative. That would connect scientific evidence to practical governance choices. A second useful format would be scenario-based workshops, for example on cross-border deployment, serious incidents, or post-deployment model updates, where participants work through what information should be shared, who should act, and where cooperation is needed. The Dialogue is already being prepared through written submissions, stakeholder consultations, and a July meeting with plenary and thematic discussions, so these more interactive formats could build naturally on the existing structure. I would also recommend fishbowl discussions, small policy labs, and regional challenge clinics. Fishbowl sessions can mix governments, researchers, companies, and affected communities in one conversation instead of separating them into stakeholder blocks. Policy labs can focus on one concrete output, such as baseline expectations for incident reporting or audit access. Regional clinics can let countries with similar capacity constraints compare needs and identify shared support options. To keep the Dialogue practical, each session should conclude with a short written note that captures areas of convergence, open disagreements, and possible next steps. That would make the process more dynamic while still producing usable outcomes.

Useful examples already exist across both AI policy and institutional design. The EU AI Act's risk-based regime for general-purpose AI models is valuable because it combines tiered obligations with practical compliance tools, including technical documentation, guidance, and the EU SEND channel for submitting required materials. The NIST AI Risk Management Framework and its Playbook offer another strong model because they translate broad principles into operational steps across the AI lifecycle through the functions Govern, Map, Measure, and Manage. The G7 Hiroshima AI Process Reporting Framework, launched through the OECD, is a useful example of international interoperability in practice. It gives organizations a common way to report how they apply the Hiroshima Code of Conduct, which can support more consistent governance across jurisdictions. UNESCO's Readiness Assessment Methodology is especially useful for countries still building AI governance capacity, because it helps assess legal, institutional, social, and technical readiness in a structured way. The OECD AI Incidents and Hazards Monitor is also valuable because it creates an evidence base from real incidents and hazards rather than relying only on abstract principles. A final approach worth adapting is fee-funded supervision in other technically complex sectors. For example, around 91.5 percent of the European Medicines Agency's 2026 budget comes from fees and charges. This suggests a practical model for beneficiary-pays AI oversight, where the firms creating the greatest supervisory burden help fund audits, technical expertise, and incident investigation. Together, these examples show that effective AI governance needs risk-based rules, shared reporting, incident evidence, capacity-building, and sustainable funding for enforcement.