Elloe AI Research Lab

Our delegation would regard the first Global Dialogue on AI Governance as successful if it establishes a credible foundation for sustained, inclusive, and action-oriented international cooperation. Success should be measured not only by the quality of the discussion, but also by the extent to which the Dialogue helps translate shared concerns into practical next steps. In that regard, three outcomes would be especially important. First, the Dialogue should contribute to a clearer common understanding of core governance objectives, including safety, security, human rights, transparency, accountability, and meaningful human oversight. Greater conceptual clarity would support cooperation across different national contexts and levels of development. Second, the Dialogue should advance a more inclusive governance ecosystem by identifying concrete measures for capacity-building. Many countries and institutions continue to face constraints in technical expertise, policy readiness, evaluation capability, and access to relevant infrastructure. A meaningful outcome would be a clearer pathway for narrowing those gaps and enabling broader participation. Third, the Dialogue should initiate a structured process for follow-up. Focused workstreams, defined responsibilities, and mechanisms for continued exchange would help ensure continuity between sessions and maintain momentum. In sum, a successful first Dialogue would strengthen international alignment, broaden effective participation, and establish a practical basis for implementation and continued cooperation.

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

Our delegation selected these priorities because they represent areas where international cooperation can provide the most immediate and practical value. Safe, secure and trustworthy AI is essential because public confidence in AI systems will depend on their reliability, resilience, and the presence of appropriate safeguards against misuse and harm. Without progress in this area, broader governance efforts will lack credibility. AI capacity-building is equally important. Many countries and institutions require stronger technical, regulatory, and institutional capabilities in order to participate meaningfully in governance processes and to benefit more equitably from advances in AI. Capacity-building is therefore central to both inclusion and implementation. Interoperability of governance approaches is also a priority. As AI systems, supply chains, and impacts increasingly operate across borders, unnecessary fragmentation in standards, regulatory expectations, and assurance practices may impede coordination and weaken collective responses to shared risks. Greater interoperability can support coherence while respecting national differences. Transparency, accountability, and human oversight remain indispensable because they help translate broad principles into operational safeguards. These elements make it more feasible to understand how systems are used, assign responsibility, enable review and redress where appropriate, and preserve human agency, particularly in higher-impact contexts. Taken together, these priorities support a governance approach that is practical, inclusive, and implementation-oriented.

Yes. In our view, several cross-cutting and emerging issues would benefit from more explicit attention. One is the growing concentration of AI capabilities across compute, data, cloud infrastructure, and advanced model development. This concentration has implications for equitable access, resilience, competition, and the ability of smaller states and institutions to participate meaningfully in both governance and innovation. A second issue is the development of stronger assurance and evaluation ecosystems. Effective governance will increasingly depend on independent testing, auditing, benchmarking, incident reporting, and post-deployment monitoring. Without such capacities, it may remain difficult to assess whether governance commitments are being implemented in practice. A third issue is environmental sustainability. As AI systems scale, governance discussions should more directly consider energy use, water consumption, hardware supply chains, and electronic waste. Finally, greater attention should be given to information integrity and content provenance, particularly in relation to synthetic media, impersonation, fraud, and the cumulative effects of AI-generated content on public trust. These issues are cross-cutting because they affect safety, development, human rights, market structure, and implementation capacity. For that reason, they merit sustained multistakeholder attention in future stages of the Dialogue.

In our sector, governance gaps across safe and trustworthy AI, capacity-building, interoperability, and transparency, accountability, and human oversight are already having material effects. The most significant challenge is uneven institutional readiness. Many organizations are under pressure to adopt AI quickly, while governance capacities have not kept pace. Risk assessment, independent evaluation, documentation, procurement standards, and internal oversight remain inconsistent. This creates uncertainty about which systems are sufficiently reliable for higher-impact use cases and increases exposure to safety, compliance, reputational, and human rights risks. A second challenge is fragmentation across jurisdictions. Divergent policy frameworks, technical standards, and assurance expectations make cross-border deployment, collaboration, and compliance more complex and costly. These burdens often fall most heavily on smaller institutions, which may lack the legal, technical, and operational resources needed to navigate multiple regimes. Capacity gaps also remain significant. Access to skilled personnel, evaluation tools, compute, and implementation guidance is uneven, limiting both responsible adoption and meaningful participation in governance processes. At the same time, these developments create important opportunities. Growing attention to safety, transparency, and human oversight is encouraging stronger governance practices, better documentation, and more disciplined deployment of AI in higher-impact contexts. Progress toward more interoperable approaches could reduce duplication, strengthen trust, and support more consistent international cooperation. Expanded capacity-building could also broaden participation, strengthen local expertise, and help ensure that the benefits of AI are more widely shared. Overall, the current moment is defined by both urgency and opportunity: the need to close governance gaps quickly, and the chance to build more inclusive, trustworthy, and effective AI ecosystems as a result.

Our delegation believes the AI Dialogue can play a distinctive role as the inclusive United Nations space for sustained international cooperation on AI governance. The Global Digital Compact and General Assembly resolution 79/325 already frame the Dialogue as a mechanism through which Member States and stakeholders can share practices, build common understanding, strengthen interoperability across governance approaches, and advance implementation of broader UN commitments on AI. Its value should lie in connecting, rather than duplicating, existing efforts. The Dialogue can help translate scientific evidence, technical experience, and policy practice into a more coherent multilateral agenda. It can elevate perspectives from developing countries, identify practical capacity-building needs, and encourage more consistent approaches to safety, accountability, human oversight, and cross-border cooperation. The Dialogue can also serve as a bridge between high-level principles and implementation. By convening governments, industry, academia, civil society, and the technical community, it can identify workable areas of convergence, support follow-up on agreed priorities, and sustain political attention as AI capabilities evolve. In that way, the Dialogue can help reduce fragmentation, strengthen legitimacy, and promote a more inclusive and practical approach to international AI governance

Our delegation considers that the AI Dialogue should build on existing work and remain complementary to it. Important reference points include the Global Digital Compact and the recommendations of the Secretary-General's High-level Advisory Body, as well as the Independent International Scientific Panel on AI, which can help anchor discussion in the wider UN framework and in evidence-based analysis. UNESCO's Recommendation on the Ethics of Artificial Intelligence and its implementation tools also provide an important normative and capacity-building foundation. The Dialogue should also connect with the OECD AI Principles and the integrated OECD/GPAI partnership, which bring policy experience, expert networks, and practical work on trustworthy AI and interoperability. The Council of Europe Framework Convention contributes a legally binding instrument grounded in human rights, democracy, and the rule of law. ITU-led processes such as AI for Good, together with related sectoral initiatives, help link governance to standards, skills, and practical implementation. Relevant ISO/IEC standards can further support organizational governance and impact assessment. The added value of the AI Dialogue would be its universal convening power and its ability to connect otherwise fragmented efforts. It can provide an inclusive forum in which Member States and stakeholders compare experience, identify gaps, elevate developing-country priorities, and encourage practical convergence without requiring identical regulatory models. Its strongest contribution would be to improve coherence, inclusiveness, and implementation across the existing AI governance landscape.

Our delegation considers that the AI Dialogue will be strongest if different stakeholders are engaged in ways that reflect their distinct roles, expertise, and responsibilities. Member States should provide policy direction, share national and regional experiences, and identify areas where greater international cooperation is needed. International organizations can contribute comparative analysis, coordination, and implementation support. The private sector and technical community can offer operational insight on model development, deployment, evaluation, standards, and risk management. Academia can provide independent research and evidence. Civil society, labour representatives, consumer groups, and affected communities can help ensure that discussions remain grounded in rights, inclusion, and real-world impacts. In terms of format, the Dialogue would benefit from combining plenary sessions with smaller, issue-focused working discussions. Plenaries are useful for shared framing and political guidance, while roundtables and breakout sessions are better suited to practical exchanges, lessons learned, and identification of areas of convergence. To improve continuity, written inputs should be invited in advance and reflected in the agenda. Short outcome summaries should be issued after each session, highlighting key points, open questions, and possible follow-up actions. The Dialogue could also include intersessional workstreams or virtual consultations on selected themes, so that progress is sustained between formal meetings. A balanced structure should preserve a central role for Member States while ensuring meaningful and well-organized stakeholder participation through moderated panels, expert evidence sessions, and targeted consultations. Hybrid access, multilingual interpretation, accessible documentation, and transparent participation criteria will be important to broaden inclusion and make engagement more effective.

In our view, several voices remain underrepresented in global discussions on AI governance. These include stakeholders from developing countries, least developed countries, small island developing States, and other contexts where policy capacity, technical infrastructure, and access to AI resources remain limited. Perspectives from low-resource language communities are also often insufficiently reflected. In addition, more consistent space is needed for workers, educators, public-interest researchers, consumer advocates, independent technologists, persons with disabilities, Indigenous Peoples, youth, and communities that are directly affected by the deployment of AI systems in public services, labour markets, education, and information environments. Smaller enterprises and institutions are also frequently underrepresented, despite the fact that fragmentation in standards, compliance expectations, and access to technical resources can affect them significantly. Their experience is important for understanding how governance frameworks function in practice beyond the largest firms and best-resourced actors. Inclusion should therefore be addressed through both representation and process design. Practical steps could include support for participation from underrepresented regions, hybrid and low-bandwidth access, multilingual interpretation and translation, accessible materials, and regional preparatory consultations that feed into the global process. Open calls for written submissions can also broaden input, especially when combined with clear guidance and transparent synthesis of contributions. It would also be useful to ensure that participation is not limited to speaking roles in plenary settings. Smaller, facilitated discussions can create more meaningful opportunities for underrepresented communities to shape agendas, identify concerns early, and contribute practical recommendations. More inclusive participation will strengthen both the legitimacy and the quality of the Dialogue.

Our delegation believes that innovative engagement formats should be designed to promote practical exchange, inclusive participation, and genuine interaction, rather than relying exclusively on formal statements. One useful format would be moderated policy labs focused on specific governance challenges, such as model evaluation, transparency measures, cross-border interoperability, or capacity-building. These smaller sessions could bring together governments, technical experts, civil society, and industry to examine concrete problems and identify areas of emerging convergence. Scenario-based discussions could also be effective. Participants could respond to realistic cases involving public-sector deployment, cross-border risks, synthetic media, or high-impact use cases. This would help move the discussion from abstract principles to operational questions of implementation, accountability, and coordination. Another valuable format would be expert evidence sessions in which researchers, practitioners, and affected communities present concise inputs followed by interactive exchange with delegations. This could strengthen the factual basis of the Dialogue and ensure that policy discussions remain connected to lived experience and technical reality. Regional dialogue hubs linked to the main process could also broaden participation and allow perspectives from different regions to be brought into the global discussion in a more structured way. Digital participation tools, if carefully moderated, may further support inclusive engagement between sessions. Overall, the most effective formats are likely to be those that combine political discussion with problem-solving, allow for interaction across stakeholder groups, and generate outputs that can inform follow-up work. The Dialogue should be both deliberative and practical, with formats that encourage listening, exchange, and concrete progress.

Our delegation sees particular value in examples that combine high-level principles with operational mechanisms. At the policy level, UNESCO's Recommendation on the Ethics of Artificial Intelligence offers a global normative framework linked to concrete policy actions, while the OECD AI Principles provide a widely used basis for trustworthy AI policy and cross-jurisdictional alignment. NIST's AI Risk Management Framework and ISO/IEC 42001 are also useful because they translate governance into organizational processes, risk management, documentation, and continuous improvement. (unesco.org) At the practice level, several measures stand out: ex ante impact assessments for higher-risk systems; clear documentation and traceability; independent testing and red-teaming; meaningful human oversight in consequential decisions; incident reporting; and post-deployment monitoring with avenues for review and remedy where harm occurs. These practices help convert general commitments into verifiable safeguards and more responsible deployment. (unesco.org) At the platform level, Singapore's AI Verify is a useful example of how governance can be made more practical through standardized testing, process checks, and reporting. Assurance-oriented sandboxes and evaluation tools are similarly valuable because they help organizations test claims about safety, robustness, transparency, and accountability before and during deployment. (imda.gov.sg) More broadly, the most promising approach is risk-based and interoperable governance: common baseline principles, stronger requirements for higher-impact uses, and standards or toolkits that can be adapted across sectors and jurisdictions. This kind of model can support innovation while improving trust, comparability, and implementation. (oecd.org)