Frontier Tech Governance Initiative

A successful first Global Dialogue on AI Governance would move beyond high-level principles to establish a shared direction for how interoperable AI governance can be operationalized in practice. An important outcome would be agreement to initiate a practical workstream such as development and provision of a sandbox focused on pluralistic alignment governance, where countries can explore how to express their legal, cultural, and societal requirements in forms that AI systems can implement while maintaining shared global safety baselines. This would directly address a central tension highlighted in the Dialogue: how to ensure AI systems are both globally interoperable and locally legitimate, without leading to fragmentation of the AI ecosystem. Such a workstream could aim to produce, over time, common approaches, templates, or reference models that enable policymakers to articulate governance requirements in interoperable ways across systems and jurisdictions. Success for the Dialogue would therefore be measured not only by consensus on principles, but by the initiation of concrete, collaborative mechanisms that translate those principles into implementable governance pathways. I would like to share a policy memo that describes the AI model alignment governance challenge: https://ssrn.com/abstract=5345186

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

My selected priorities reflect a common underlying concern: how to ensure that AI systems are both safe and trustworthy, while remaining interoperable across jurisdictions and responsive to diverse social, economic, and cultural contexts. These areas are closely interdependent. Safety and trust cannot be sustained if AI systems do not reflect the legal frameworks and societal norms of the environments in which they are deployed. At the same time, approaches that are overly fragmented risk undermining interoperability and limiting equitable access to AI capabilities. That is the alternative where states respond by attempting to build fully separate AI ecosystems aligned only to their own norms, would lead to fragmentation of the global AI landscape. Such fragmentation would undermine interoperability, reduce access to leading technologies, and ultimately weaken the shared trust that safe AI systems require. From my perspective, this points to the need for governance models that reconcile global safety standards with pluralistic, context-sensitive alignment. Advancing such models would help ensure that AI systems are not only technically robust, but also socially legitimate, supporting more inclusive and beneficial adoption across different regions and communities.

One cross-cutting issue that may warrant greater attention is the operationalization of AI governance, essentially how policy objectives, legal requirements, and societal norms are translated into the actual behavior of AI systems. Many of the listed themes such as safety, trustworthiness, interoperability, and broader societal impacts, depend not only on well-articulated principles, but on the ability, the frameworks and incentives to implement those principles consistently across different systems and jurisdictions. At present, there remains a gap between policy-level intent and system-level execution, particularly in how alignment requirements are specified, tested, and enforced in practice. Other domains offer useful precedents. In internet governance, mechanisms such as ICANN have enabled coordinated interoperability across globally distributed systems; in telecommunications, ITU standards have provided common technical frameworks; and in broader technology governance, ISO standards have helped translate policy objectives into auditable and implementable practices. A similar approach needs to be established for AI governance, which could also form the basis for focused investigation by the UN's Independent International Scientific Panel on AI, helping to develop evidence-based approaches to making AI governance implementable, interoperable, and measurable in real-world deployments.

AI diffusion in Africa is accelerating, but governance still lacks tools to translate productivity gains into net jobs, risking job‑poor growth amid labor‑force expansion and ~83% informality. The specific challenges can be framed as: - Limited task/sector modelling; no clear requirement to assess how AI affects jobs - Informal/service delivery often outside policy instruments In a reference example, the African Development Bank expects AI could add $1T to Africa's GDP by 2035 and support ~35–40M jobs (African Development Bank, Africa's AI Productivity Gain: Pathways to Labour Efficiency, Economic Growth and Inclusive Transformation, 2025), yet employment is commonly inferred from historical GDP–employment elasticities rather than explicit sector pathways. The African Union Continental AI Strategy notes limited evidence on AI's socio‑economic impacts in Africa and warns of an AI divide from gaps in data, compute, and talent. The World Bank warns growth is insufficient for jobs: only ~24% of new workers land wage‑paying jobs, and up to 12M youth enter the labour market each year versus ~3M new formal wage jobs created annually. Without labour‑sensitive procurement and transition rails, deployments may default to cost‑cutting automation in formal sectors (finance, public administration), intensifying underemployment and inequality. Divergent national restrictions could also fragment AI governance, reducing interoperability. Comparative opportunity: With fewer entrenched legacy workflows, Africa can embed "human-in-the-loop at scale" from the outset, ensuring for instance that AI is deployed to expand healthcare access and throughput, with attention to how tasks are redistributed across the workforce. I think there is opportunity to invite the UN AI Scientific Panel to scope and pilot an "AI‑jobs governance module" (metrics + augmentation‑first procurement templates) as input to an interoperable policymaker toolkit.

The AI Dialogue can add the most value by supporting rails that translate principles to practical, testable cooperation mechanisms. First, it can convene focused working sessions that are co-led by the UN AI Dialogue Secretariat in collaboration with bodies such as the OECD, UNESCO, and industry coalitions like the Frontier Model Forum, to translate high-level commitments into implementable tools (e.g., safety baselines, evaluation protocols, reporting templates). These sessions should produce draft artefacts, not just summaries. Second, the Dialogue can trigger the establishment of "interoperability labs", with technical coordination from the International Telecommunication Union (ITU) and regional bodies such as Smart Africa, to test how different national approaches can align in practice (e.g., safety assessments, data governance, labor-impact measurement), generating outputs that can be adopted across jurisdictions. Third, it can establish a standing collaboration channel with standards bodies, notably the International Organization for Standardization (ISO), International Electrotechnical Commission (IEC), and the ITU, to translate governance approaches into technical standards that enable interoperability and scale. Fourth, the Dialogue can support post-session implementation tracks, potentially anchored through UN system platforms such as the UNDP, UNESCO and UNCTAD, where countries and partners continue to refine and deploy these artefacts. Finally, adopting an iterative "build-and-test" model of engagement, rather than one-off consultations, can accelerate convergence by grounding cooperation in working implementations. In this way, the Dialogue can function not only as a forum for exchange, but as a platform for co-developing and operationalizing interoperable AI governance.

Several existing initiatives provide strong foundations for international cooperation on AI governance, but remain insufficiently connected and operationalized. Normative frameworks advanced by the OECD and UNESCO have helped establish shared principles for trustworthy AI. Standards bodies including ISO/IEC and the ITU are translating these into technical specifications. Industry-led efforts such as the Frontier Model Forum are advancing safety practices, while academic and research institutions are contributing critical work on evaluation, alignment, and impact assessment. Regional platforms such as Smart Africa and the recently launched Africa AI Council are working to align continental priorities with national implementation. However, these efforts often operate in parallel rather than in coordination, and a gap remains between high-level principles and deployable, interoperable governance mechanisms. The AI Dialogue can add value by acting as a bridging layer across these communities. Specifically, it can: - connect normative frameworks with standards development processes, - align industry practices with public governance objectives, - integrate academic research into policy and implementation pathways (via the Scientific Panel), - and support regional adaptation through structured collaboration. Beyond convening, the Dialogue can introduce practical coordination mechanisms such as shared artefacts, interoperability testbeds, and ongoing implementation tracks, that enable convergence in practice. In doing so, the Dialogue can shift the ecosystem from fragmented efforts to a more coherent, execution-oriented model of international cooperation, where principles, standards, and implementation evolve together.

Different stakeholders can contribute most effectively to the AI Dialogue when engagement is structured around complementary roles in a shared build-and-test model, rather than general consultation. Governments can articulate policy objectives and constraints, including safety, accountability, and socio-economic priorities. Standards bodies (ISO/IEC, ITU) can translate these into technical specifications and interoperability frameworks. Industry can contribute implementation experience, safety practices, and operational data. Academia provides evaluation methods, impact analysis, and independent validation, while civil society and policy practitioners can surface societal impacts, rights considerations, and practical governance design insights, ensuring that frameworks remain grounded, inclusive, and implementable. Regional platforms can support contextual adaptation and cross-border coordination. To enable this, the Dialogue could adopt a three-part structure: 1. Focused Working Sessions: Multi-stakeholder groups tasked with producing draft artefacts (e.g., safety baselines, reporting templates, labor-impact metrics). 2. Interoperability Tracks: Ongoing collaboration triggered through the Dialogue and coordinated with bodies such as ITU and ISO, to test alignment of governance approaches in practice. 3. Implementation Pathways: Post-session tracks, supported by UN entities (e.g., UNDP, UNCTAD), to refine and deploy outputs. Incorporating technical and policy clinics where research, field experience, and policy design are jointly stress-tested can further strengthen outcomes. This structure shifts participation from representation to co-creation, enabling stakeholders to contribute working components of interoperable AI governance.

A key underrepresented perspective in global AI governance discussions is that of end-users and affected individuals, particularly those whose interaction with AI systems is indirect or mediated through services (e.g., workers, patients, small businesses). While stakeholder representation often includes governments, industry, and civil society organizations, there remains limited systematic incorporation of lived experience at scale, i.e. how AI systems actually shape access, opportunity, and outcomes in everyday contexts. The Dialogue could address this by introducing structured input mechanisms such as: - aggregated user feedback channels embedded in deployed AI services, - representative "user panels" or citizen assemblies linked to specific use cases, and - partnerships with regional platforms to capture localized insights. Incorporating such inputs in a standardized, comparable format would allow user-level impacts to inform governance alongside institutional perspectives.

Meaningful engagement in the AI Dialogue will require moving beyond static interventions toward structured, participatory, and feedback-driven formats. One approach is to incorporate aggregated user input channels, where insights from individuals interacting with AI systems across sectors such as health, finance, and public services, are synthesized and presented as structured evidence during sessions. This enables the Dialogue to engage not only institutional perspectives, but lived experience at scale. Second, the Dialogue could convene use-case-specific citizen panels or practitioner forums, composed of workers, service providers, and small enterprises directly affected by AI deployment. These panels can present distilled insights on how AI is reshaping tasks, roles, and access in practice. Third, pre-Dialogue policy and technical clinics can be organized with policymakers, researchers, engineers, and practitioners to stress-test governance approaches against real-world scenarios. The outputs such as refined proposals, identified gaps, and tested assumptions, can then be presented at the Dialogue as decision-ready inputs, rather than discussed from first principles. Together, these formats shift engagement from representation to evidence-informed co-creation, enabling the Dialogue to operate as a point of convergence for work already grounded in practice.

Several emerging approaches offer practical pathways toward more effective AI governance, particularly where they bridge principles with implementation. A strong example is the growing role of international standards, reinforced at the International AI Standards Summit (Seoul, 2025), where the ISO, IEC, and ITU jointly launched the Seoul Statement, positioning standards as the primary mechanism for translating policy principles into interoperable, real-world systems. Frameworks such as ISO/IEC 42001 and 23894 exemplify this by embedding governance into operational processes. Second, soft-law mechanisms are already shaping convergence. Examples include the G7 Hiroshima AI Process and voluntary safety commitments by leading AI developers, which establish shared expectations on model evaluation, transparency, and risk mitigation ahead of formal regulation. Third, polylateral governance models such as Internet coordination through ICANN and global telecommunications standardization via the ITU demonstrate how multi-stakeholder systems can sustain interoperability across jurisdictions while accommodating diverse policy environments. Building on these, multi-layered governance approaches are being explored to address the translation gap between policy and implementation in enabling jurisdictions to express legal and societal requirements in structured, system-level forms while maintaining cross-border interoperability (e.g., Multi-Layered Alignment Governance for AI Models - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5345186). Finally, there is increasing focus on deployment-level impact mechanisms. Examples include Canada's Algorithmic Impact Assessment (AIA) and Singapore's AI Verify testing and validation framework, both of which structure how risks are assessed and managed prior to deployment. In parallel, some regulatory approaches are beginning to link compliance to recognized standards. For example, safe harbor provisions under the Texas Responsible AI Governance Act (TRAIGA) recognize compliance with frameworks such as the NIST AI Risk Management Framework. There are also sector-specific requirements emerging. In healthcare, California AB 3030 introduces obligations around human oversight and disclosure in AI-assisted decision-making. Together, these approaches anchor governance directly in real-world use.