University of York

First, success would involve establishing a shared, operational understanding of interoperability. Rather than seeking legal harmonisation, the Dialogue should clarify how different national and regional frameworks can perform equivalent governance functions—such as risk assessment, accountability allocation, monitoring, and redress—thereby enabling coordination without constraining regulatory sovereignty. Second, the Dialogue would be successful if it identifies a global minimum governance baseline for high-impact AI systems. This should include core elements such as proportionate risk assessment, lifecycle oversight, transparency requirements, and mechanisms for contestability. Agreement at this functional level would provide a foundation for more coherent global governance while remaining adaptable to different institutional contexts. Third, success requires a clear commitment to capacity-building as a central pillar of governance. This includes support for regulators and public institutions in developing countries, through shared methodologies, technical expertise, and practical tools for evaluation and oversight. Fourth, the Dialogue should clarify the division of roles between the Global Dialogue and the scientific panel established under the United Nations General Assembly framework, ensuring a strong link between evidence generation and policy translation. Finally, a successful outcome would include tangible outputs, such as a summary of convergence areas, a repository of best practices, and a forward-looking agenda. These would ensure continuity and demonstrate that the Dialogue delivers actionable value rather than remaining purely deliberative.

• Interoperability of governance approaches
• AI capacity-building
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

1. Interoperability of governance approaches This is central to your argument on moving from fragmented regimes to functionally aligned governance systems. Your emphasis on "governance functions" rather than legal harmonisation fits precisely here and addresses a core gap in current global efforts. 2. AI capacity-building You explicitly position capacity as core governance infrastructure, not a secondary issue. This is a high-impact area where your contribution (methods, templates, institutional design logic) would be both practical and policy-relevant. 3. Transparency, accountability, and human oversight These are key operational pillars within your proposed global minimum governance stack, particularly around lifecycle governance, auditability, and contestability mechanisms. 4. Social, economic, ethical, cultural, linguistic and technical implications of AI This captures your broader systems-level perspective, including: labour market restructuring, organisational transformation, systemic and cross-border risks, cognitive and behavioural implications.

First, institutional and regulatory capacity asymmetries remain insufficiently foregrounded. Differences in technical expertise, enforcement capability, and public-sector readiness fundamentally shape how AI governance is implemented in practice. Without addressing these asymmetries, even well-designed frameworks risk uneven application and exclusion from effective participation. Second, political economy and market concentration represent a critical but under-specified issue. The global AI ecosystem is characterised by concentration in compute infrastructure, data access, and frontier model development. This creates dependencies for smaller states and limits the practical autonomy of regulators. Governance discussions should therefore explicitly engage with issues of access, dependency, and fair participation in the AI value chain. Third, public-sector AI and procurement governance deserves greater prominence. Governments are increasingly major adopters of AI systems in areas such as education, welfare, and public administration. Procurement practices represent a key leverage point for embedding accountability, transparency, and safeguards, yet this dimension is often treated as secondary. Fourth, lifecycle governance of AI systems cuts across multiple thematic areas but is not explicitly identified. Risks emerge at different stages-design, training, deployment, and post-deployment adaptation-and require continuous oversight rather than point-in-time regulation. Finally, systemic and cross-border risks extend beyond individual harms. These include correlated failures, information integrity challenges, labour market restructuring, and cascading effects across interconnected systems. Such risks require coordinated international responses and cannot be adequately addressed through isolated national frameworks. Addressing these cross-cutting issues would strengthen the coherence, inclusiveness, and practical effectiveness of the Global Dialogue on AI Governance.

First, fragmentation and limited interoperability of governance approaches create uncertainty for organisations operating across jurisdictions. In the UK, the emerging sectoral and principles-based approach must increasingly interact with more prescriptive regimes (e.g., the EU). This generates compliance complexity, particularly for firms deploying AI across borders, and complicates the design of consistent risk management and accountability structures. Second, capacity constraints are becoming evident at the institutional level. While the UK has strong research capability, there are gaps in regulatory and organisational capacity to evaluate, procure, and monitor AI systems effectively. In universities and public institutions, this manifests as uneven adoption, limited auditing capability, and reliance on external vendors without fully developed oversight mechanisms. Third, gaps in transparency, accountability, and human oversight are increasingly salient in applied settings. In finance and algorithmic decision-making contexts, the use of complex models raises challenges around explainability, validation, and responsibility allocation. Similarly, in education, the rapid uptake of generative AI tools has outpaced the development of clear governance standards, creating risks related to academic integrity, assessment design, and cognitive dependency. Fourth, the social and economic implications of AI are already observable through shifts in skill demand, changes in work organisation, and increasing reliance on automated decision systems. These developments are unevenly distributed, raising concerns about inequality and institutional preparedness. Overall, these governance gaps do not merely pose theoretical risks; they are actively shaping operational practices, institutional resilience, and policy effectiveness across sectors. Addressing them requires coordinated efforts that combine regulatory clarity, capacity-building, and practical governance tools.

First, the Dialogue can advance cooperation by promoting interoperability across governance approaches. By focusing on shared governance functions—such as risk assessment, accountability, monitoring, and redress—it can enable alignment between diverse regulatory systems without requiring legal harmonisation. This would reduce fragmentation while preserving national policy autonomy. Second, the Dialogue can serve as a mechanism for linking scientific evidence to policy practice. In coordination with the scientific panel established under the United Nations General Assembly framework, it can translate technical insights into feasible and context-sensitive governance approaches. This strengthens evidence-based policymaking and reduces the gap between technical developments and regulatory responses. Third, the Dialogue can facilitate capacity-building and inclusive participation, particularly for developing countries and under-resourced institutions. By supporting knowledge exchange, shared methodologies, and access to expertise, it can help ensure that all countries are able to engage meaningfully in AI governance and implementation. Fourth, the Dialogue can provide a forum for addressing systemic and cross-border risks—including concentration of technological capabilities, supply chain dependencies, and global diffusion of high-risk systems—which cannot be effectively managed at the national level alone. Finally, the Dialogue can enhance cooperation through practical outputs, including repositories of best practices, identification of convergence areas, and forward-looking agendas. By generating actionable guidance rather than purely declarative statements, it can become a central mechanism for sustained and effective international coordination on AI governance.

Key initiatives include the OECD AI Policy Observatory and OECD AI Principles, which provide widely adopted normative guidance; UNESCO's Recommendation on the Ethics of Artificial Intelligence, which anchors AI governance in human rights and ethical standards; the G7 Hiroshima AI Process, which advances discussions on generative AI governance among major economies; and technical standard-setting efforts led by bodies such as the International Organization for Standardization and International Telecommunication Union. In addition, regional regulatory frameworks—most notably the EU's AI Act—are shaping emerging compliance architectures with global spillovers. Despite these developments, current efforts remain institutionally and geographically fragmented, with limited mechanisms for coordination, comparison, and mutual learning across regimes. The added value of the Global Dialogue lies in its ability to function as a bridging and integration platform. First, it can connect normative frameworks, technical standards, and regulatory approaches by identifying common governance functions and areas of convergence. Second, it can facilitate interoperability, enabling different systems to align in practice without requiring uniform regulation. Third, it can strengthen the evidence–policy interface by linking the scientific panel's work to real-world governance challenges. Fourth, it can promote inclusive participation, particularly by amplifying the voices of countries and stakeholders not fully represented in existing initiatives. Finally, it can generate practical outputs, such as repositories of best practices and capacity-building priorities, thereby translating dispersed knowledge into actionable guidance. In doing so, the Dialogue can enhance coherence, reduce duplication, and support more effective international cooperation on AI governance.

Different stakeholders contribute distinct forms of expertise. Governments provide regulatory authority and implementation experience; the private sector contributes technical capability, deployment knowledge, and risk management practices; academia offers independent analysis, methodological rigour, and evidence synthesis; civil society brings perspectives on rights, inclusion, and societal impact; and international organisations support coordination, standard-setting, and capacity-building. The Dialogue should be designed to integrate these inputs rather than treat them as parallel streams. In terms of structure, the Dialogue would benefit from a multi-layered format. Plenary sessions can set strategic direction and identify areas of convergence, while thematic working groups—aligned with priority areas such as interoperability, capacity-building, and accountability—can develop more detailed, practice-oriented outputs. These groups should operate continuously between annual meetings to ensure continuity and progress. The Dialogue should also include structured evidence inputs, drawing on the scientific panel established under the United Nations General Assembly framework, alongside stakeholder submissions and case-based evidence from real-world deployments. This would strengthen the evidence–policy interface. To enhance inclusiveness, the format should enable balanced participation, including support mechanisms for underrepresented countries and stakeholders. Hybrid participation, regional consultations, and open submission channels can broaden engagement. Finally, the Dialogue should produce clear, recurring outputs, such as summaries of convergence, identified gaps, best practice repositories, and forward-looking agendas. A structured, iterative format would ensure that stakeholder contributions are cumulative, comparable, and translated into actionable governance insights.

First, low- and middle-income countries, particularly across Africa, parts of Asia, and small island developing states, are underrepresented. These countries are often affected by AI systems developed elsewhere but lack meaningful participation in shaping governance frameworks. Inclusion can be strengthened through targeted support for participation, regional consultations, and sustained capacity-building initiatives. Second, public-sector practitioners—including regulators, procurement officials, and frontline administrators—are not consistently represented. Yet they are central to implementation. Their inclusion could be enhanced through dedicated practitioner forums, case-based sessions, and structured input on operational challenges. Third, workers and labour representatives are underrepresented, despite significant implications of AI for employment, skills, and workplace organisation. Greater engagement with trade unions and worker associations would improve understanding of labour market impacts and transition needs. Fourth, Global South research communities and institutions often lack visibility in global AI debates. Supporting research collaboration, access to funding, and inclusion in expert panels would strengthen epistemic diversity and reduce reliance on a narrow set of knowledge producers. Fifth, civil society organisations focusing on non-Western cultural, linguistic, and social contexts remain underrepresented. This limits the ability to address culturally specific impacts of AI systems. To address these gaps, the Global Dialogue convened by the United Nations should adopt inclusive participation mechanisms, including financial and logistical support, multilingual engagement, open calls for contributions, and structured integration of stakeholder inputs into decision-making processes. Ensuring that participation translates into influence is essential for the legitimacy and effectiveness of global AI governance.

First, scenario-based policy labs could be introduced. These would bring together mixed stakeholder groups to work through realistic AI governance challenges—such as cross-border deployment of high-risk systems or public-sector procurement decisions—allowing participants to test governance approaches in practice rather than discuss them abstractly. Second, structured "evidence-to-policy" sessions should be incorporated, linking directly to the scientific panel established under the United Nations General Assembly framework. In these sessions, technical findings would be presented alongside facilitated discussions on policy implications, ensuring a clear translation from evidence to governance options. Third, comparative governance roundtables could enable jurisdictions to present their approaches in a standardised format (e.g., risk classification, accountability mechanisms, enforcement tools), followed by moderated comparison. This would support interoperability by focusing on functional similarities rather than legal differences. Fourth, case-based practitioner forums should be included, where public-sector officials, regulators, and industry practitioners present real-world implementation challenges. This would ground the Dialogue in operational realities and highlight gaps between policy design and practice. Fifth, multi-stakeholder co-creation sessions could be used to develop tangible outputs—such as draft guidance, assessment templates, or best-practice frameworks—during the Dialogue itself. Finally, continuous engagement mechanisms, including virtual working groups and open consultation platforms between annual meetings, would ensure that participation is sustained and cumulative. Together, these formats would transform the Dialogue into an interactive, outcome-oriented process rather than a purely deliberative forum.

The European Union AI Act provides a risk-based regulatory framework that classifies AI systems by level of risk and links obligations-such as conformity assessments, documentation, and oversight-to those risk tiers. This approach operationalises proportionality and lifecycle governance. The OECD AI Policy Observatory supports policy coordination and benchmarking, offering tools and comparative data that enable governments to align governance approaches while retaining flexibility. UNESCO's Recommendation on the Ethics of Artificial Intelligence provides a globally endorsed normative framework, linking AI governance to human rights, sustainability, and social inclusion, and supporting implementation through readiness assessments. In the United Kingdom, the UK Government has adopted a principles-based, sector-led approach, supported by guidance from regulators. Complementary tools such as algorithmic transparency records in the public sector demonstrate practical mechanisms for accountability. Technical standard-setting bodies, including the International Organization for Standardization and International Telecommunication Union, contribute by developing auditable standards and management systems, enabling organisations to operationalise governance requirements. Emerging practices also include AI impact assessments, model documentation (e.g., model cards), and post-deployment monitoring frameworks, which embed accountability across the lifecycle of AI systems. Collectively, these examples illustrate that effective AI governance combines risk-based regulation, international coordination, technical standards, and practical implementation tools. The added value of the Global Dialogue lies in connecting these approaches, identifying transferable elements, and supporting their adaptation across diverse institutional contexts.