Strategy ARX Advisory Limited

A successful first Global Dialogue on AI Governance would produce outcomes that move beyond principle-setting into system-level alignment between governance, infrastructure, and economic reality. First, it should establish a shared baseline understanding that AI is not only a technological system, but a tightly coupled socio-technical and economic system, where outcomes are shaped by interactions across models, networks, cloud infrastructure, and regulatory frameworks. Second, it should progress from high-level principles to operational governance mechanisms, including: common definitions of AI system reliability (beyond model performance alone) early interoperability frameworks for cross-border AI systems shared approaches to certification of AI infrastructure and services Third, it should explicitly connect access and capacity-building to system performance, not just availability of tools. Bridging AI divides requires not only access to compute and models, but also the ability to operate AI systems within stable, low-latency, and predictable infrastructure environments. Fourth, it should recognise that value creation in AI systems is increasingly determined by interaction outcomes rather than data or compute volume alone, and that current economic and infrastructure models may not reflect this shift. Finally, a meaningful outcome would be agreement on a roadmap for continued collaboration between states, industry, and multilateral institutions to ensure that AI governance evolves in step with both system dynamics and emerging economic structures, not independently of them.

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

These four priorities reflect the need to align AI governance with both system behaviour and infrastructure realities, rather than treating governance as separate from technical and economic design. Safe, secure and trustworthy AI is fundamental because AI systems are increasingly embedded in critical infrastructure and economic systems. Trustworthiness must extend beyond model outputs to include end-to-end system reliability across interconnected environments. AI capacity-building is essential to ensure equitable participation in the AI economy. However, capacity must be understood broadly to include not only access to tools and compute, but also the ability to operate AI systems within performant, resilient digital infrastructure. Interoperability of governance approaches is increasingly important as AI systems operate across borders and jurisdictions. Without alignment, fragmented governance risks reducing effectiveness and increasing systemic inefficiencies in cross-border AI deployment. Transparency, accountability, and human oversight are critical given the complexity of AI systems, where outcomes are shaped not only by models but also by infrastructure behaviour, orchestration layers, and system-level interactions. Accountability frameworks must therefore extend across the full stack. Collectively, these priorities reflect the need to govern AI as a system-of-systems, where technical performance, economic incentives, and regulatory frameworks are tightly interdependent.

A key cross-cutting issue not fully captured in the listed themes is the role of system-level economics and infrastructure behaviour in shaping AI outcomes. Current governance discussions appropriately focus on safety, trust, access, and rights. However, AI systems also operate as tightly coupled economic and technical systems, where small variations in latency, reliability, and orchestration can propagate across networks and materially influence outcomes. This introduces an emerging challenge: AI value creation is increasingly determined by interaction-level performance rather than traditional measures of compute usage, data access, or bandwidth consumption. As a result, existing economic and infrastructure models may not fully reflect where cost is incurred or where value is captured. A second emerging issue is the rise of hidden systemic effects, such as retry amplification, load propagation, and cross-layer dependencies between networks, cloud systems, and AI applications. These effects may not be visible at the level of individual components but can significantly impact overall system stability and fairness. A third issue is the growing importance of control points in AI systems, including standards, certification regimes, and interoperability frameworks. These mechanisms increasingly determine how AI systems behave across jurisdictions and therefore where value and influence concentrate. Addressing these gaps requires complementing existing governance frameworks with a stronger focus on system dynamics, infrastructure economics, and interaction-level performance measurement, ensuring that governance evolves in step with how AI systems actually operate in practice.

Governance gaps in AI are increasingly manifesting at the intersection of infrastructure capability, system economics, and cross-border interoperability, with direct implications for national and sectoral competitiveness. One of the most significant challenges is the misalignment between traditional governance assumptions and the actual behaviour of AI systems. Current frameworks tend to treat AI as a software layer, while in practice it operates as a tightly coupled system spanning networks, cloud infrastructure, models, and enterprise workflows. This creates governance blind spots where system-level effects - such as latency sensitivity, retry amplification, and orchestration overhead - are not adequately reflected in policy or economic design. A second challenge is the emergence of hidden value and cost structures. AI workloads generate costs and performance impacts that are not fully captured in conventional pricing or regulatory models, which are still largely based on bandwidth, compute, or data usage. This can result in structural inefficiencies and unintended value transfer across the ecosystem. A third challenge is the fragmentation of governance approaches across jurisdictions, which increases complexity for cross-border AI deployment and limits interoperability of systems, standards, and compliance frameworks. At the same time, these developments create significant opportunities. Countries and regions that invest in sovereign AI infrastructure capabilities - including connectivity, compute, and AI service layers - are better positioned to participate in AI value creation rather than solely consume external services. Similarly, the evolution of standards, certification regimes, and interoperability frameworks presents an opportunity to shape how AI systems operate globally. Finally, there is an opportunity to strengthen governance by integrating system-level performance and economic dynamics into regulatory thinking, ensuring that policy reflects how AI actually functions in operational environments. Overall, these gaps highlight the need for a more integrated approach linking governance, infrastructure, and system economics.

The AI Dialogue can play a critical role in shifting international cooperation from principle-based alignment to system-aware governance coordination. At present, global AI governance discussions are often fragmented across safety, innovation, access, and rights agendas. The Dialogue provides an opportunity to integrate these dimensions into a more coherent framework that reflects the reality that AI operates as a tightly coupled socio-technical and economic system, spanning infrastructure, models, cloud platforms, and regulatory environments. A key contribution of the Dialogue could be to establish a shared understanding that effective AI governance requires alignment across three interdependent layers: - Governance frameworks (trust, accountability, human rights) - Infrastructure systems (connectivity, compute, interoperability) - System economics (how value and cost are created through AI interactions) This would enable more practical cooperation on issues such as interoperability of regulatory approaches, cross-border reliability standards, and shared definitions of trustworthy AI system behaviour. The Dialogue can also serve as a platform to move toward common operational concepts, including how to define and measure AI system reliability beyond model performance alone, and how to ensure that governance frameworks reflect real-world deployment conditions. Importantly, it can help bridge the gap between developed and developing economies by connecting AI governance to capacity-building in digital infrastructure and system-level capability, rather than treating access as purely a question of tools or models. Ultimately, the Dialogue's value lies in enabling a shift from fragmented governance discussions to a coordinated global approach that reflects how AI systems actually function and scale in practice.

The AI Dialogue should build on a range of existing international and multi-stakeholder initiatives spanning AI safety, digital governance, and infrastructure development, while providing a unifying layer that connects them. Relevant foundations include: - Emerging AI safety and governance frameworks developed by multilateral organisations and national governments - Digital cooperation initiatives focused on connectivity, data governance, and digital public infrastructure - Standards development bodies working on interoperability, cybersecurity, and system reliability - Open-source and open-data initiatives supporting broader access to AI tools and models - Capacity-building programmes aimed at strengthening digital and computational capabilities in developing economies While these initiatives address important components of the AI ecosystem, they often operate in relative isolation - focused either on governance principles, infrastructure access, or technical standards. The added value of the AI Dialogue would be to provide a system-level coordination platform that explicitly links these domains. In particular, it can help: - Align governance frameworks with the operational realities of AI systems, including their infrastructure dependencies and performance dynamics - Connect capacity-building efforts with end-to-end AI system capability, including compute, networking, and deployment environments - Support convergence on shared concepts such as interoperability, system reliability, and cross-border AI functionality - Identify and address gaps where economic and system-level behaviours of AI are not fully reflected in existing governance or standards work By doing so, the AI Dialogue can move beyond aggregation of existing efforts toward enabling coherent global coordination across governance, infrastructure, and system economics, ensuring that AI development is both scalable and aligned with public interest objectives.

Different stakeholders can contribute most effectively to the AI Dialogue by engaging across complementary layers of the AI ecosystem - governance, infrastructure, and system behaviour. - Member States can define policy priorities, regulatory principles, and sovereignty considerations, particularly around safety, rights, and national capability development. - Industry and infrastructure providers can contribute practical insights on deployment realities, system constraints, interoperability challenges, and evolving cost structures in AI systems. - Technical and standards bodies can support convergence on shared definitions of system reliability, interoperability, and security. - Academia and civil society can provide independent analysis on societal impacts, accountability, and human rights implications. - Developing country stakeholders can highlight capacity gaps and structural barriers to participation in AI-driven economies. To ensure meaningful engagement, the AI Dialogue should adopt a layered structure: 1. Strategic plenary discussions focused on high-level governance principles 2. Technical and operational working groups addressing interoperability, infrastructure, and system reliability 3. Applied thematic tracks linking AI governance to sectoral outcomes (e.g., health, finance, public services) 4. Capacity-building forums specifically designed for emerging economies and underrepresented stakeholders This structure would allow the Dialogue to move beyond general principles toward actionable alignment between governance, infrastructure, and real-world system behaviour.

Several key perspectives remain underrepresented in global AI governance discussions. First, developing and emerging economies are often insufficiently represented in shaping AI system design and standards, despite being significantly affected by their deployment. Their inclusion is essential to ensure that governance frameworks reflect diverse infrastructure realities and capacity constraints. Second, infrastructure operators and network-level stakeholders are frequently underrepresented, even though AI systems depend heavily on underlying connectivity, compute, and orchestration environments. Their insights are critical to understanding system reliability, scalability, and real-world constraints. Third, technical operational perspectives focused on system behaviour - including latency, resilience, and cross-layer dependencies - are often missing from high-level policy discussions, despite being central to AI system performance and outcomes. Fourth, civil society voices from the Global South, particularly those focused on digital equity, labour impacts, and access to digital infrastructure, are underrepresented in shaping global standards and governance frameworks. To address these gaps, inclusion should be strengthened through: - structured regional representation in all thematic tracks - dedicated technical-infrastructure advisory groups - funded participation mechanisms for developing countries - integration of civil society and academic voices into working groups, not only plenaries - hybrid participation models enabling broader accessibility A more balanced governance process requires ensuring that those who operate, depend on, and are structurally affected by AI systems are directly involved in shaping their rules and evolution.

To foster meaningful engagement, the AI Dialogue should adopt formats that reflect the system-level nature of AI, rather than treating it solely as a policy discussion. First, scenario-based system simulations could be used to explore how AI governance decisions propagate through interconnected infrastructure, economic, and societal systems. This would help participants understand second-order effects, such as how small changes in system design can amplify across networks and services. Second, cross-disciplinary "stack tables" could be introduced, bringing together policymakers, infrastructure operators, technologists, and civil society in structured discussions aligned to the AI stack (applications, models, cloud, networks, governance). This would ensure alignment across layers rather than siloed debate. Third, live technical-policy translation sessions could bridge gaps between technical system behaviour (e.g. latency, interoperability, reliability) and governance concepts such as trust, safety, and accountability. Fourth, regional deep-dive labs could allow focused engagement on local infrastructure constraints, capacity-building needs, and implementation realities, particularly for developing economies. Finally, a continuous digital collaboration platform could extend the Dialogue beyond formal meetings, enabling iterative input, shared documentation, and evolving consensus on standards and frameworks. Together, these formats would shift the Dialogue from a static consultation process to a dynamic coordination mechanism, capable of reflecting the complexity of AI as a system-of-systems while enabling practical, inclusive, and actionable outcomes.

Effective AI governance is increasingly emerging from a combination of policy frameworks, technical standards, and infrastructure-level practices that together address both the governance of AI systems and the systems in which AI operates. One key example is the development of risk-based AI regulatory frameworks, which classify AI systems according to their potential impact and apply proportionate governance requirements. These approaches help ensure that oversight is aligned with system criticality, rather than applying uniform rules to all use cases. Another important mechanism is the evolution of technical standards for AI safety, security, and interoperability, developed through international standards bodies. These provide a foundation for aligning system behaviour across jurisdictions, particularly in areas such as model evaluation, cybersecurity, and system reliability. In addition, digital public infrastructure models are increasingly relevant, particularly where governments provide foundational digital services that enable scalable and inclusive access to AI-enabled systems. These approaches support broader participation in the AI economy while maintaining governance oversight. From an infrastructure perspective, emerging practices around sovereign AI and trusted digital infrastructure demonstrate how nations can maintain control over data, compute, and AI services while still participating in global innovation ecosystems. This includes certification-based approaches to infrastructure trust and compliance. Open-source ecosystems and open AI model initiatives also play a significant role by improving transparency, enabling independent evaluation, and supporting broader access to AI capabilities, particularly in resource-constrained environments. Finally, multi-stakeholder platforms that bring together governments, industry, technical bodies, and civil society are essential for ensuring that governance frameworks remain aligned with both system-level realities and societal expectations. Collectively, these approaches highlight the importance of integrating policy, technical standards, and infrastructure design to create AI governance systems that are both practical and adaptable to rapidly evolving technological conditions.