Middle States Association | OECD AI Governance Task Force (G7 Hiroshima AI Process)

A successful first Global Dialogue on AI Governance should move beyond principle-setting toward operational alignment at the level where institutions make consequential decisions. Three outcomes would define success: First, the establishment of a shared understanding that AI governance must extend beyond system performance and lifecycle management to include decision-level accountability. Current frameworks largely govern how systems are built and deployed; however, institutional risk materializes when human actors rely on AI outputs in real-world decisions. The Dialogue should formally recognize this governance gap. Second, the development of interoperable reference points for defining when and how AI outputs may be relied upon in practice. This includes clarifying roles, verification thresholds, and documentation requirements across different risk contexts. Without such alignment, fragmentation across jurisdictions will continue to undermine both trust and implementation. Third, the initiation of a structured pathway toward operational guidance—potentially through voluntary reporting mechanisms or pilot implementations—that translates high-level principles into enforceable institutional practices. This should include mechanisms for auditability, challenge, and accountability in AI-assisted decisions. Success, therefore, is not measured by consensus on abstract principles, but by progress toward governing the conditions under which AI is actually used in consequential decision-making.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These priorities reflect the need to bridge the gap between high-level governance principles and institutional decision-making practices. Interoperability is essential because the current landscape-spanning OECD, UN, regional regulations, and technical standards-risks fragmentation without alignment at the operational level. Institutions require coherent guidance that can be implemented across jurisdictions without duplicative or conflicting requirements. Transparency, accountability, and human oversight are central because they define how responsibility is exercised when AI informs or influences decisions. However, these concepts must be translated into concrete mechanisms, such as verification thresholds, documentation protocols, and review processes. Safe, secure, and trustworthy AI remains foundational, but trust cannot be achieved solely through system-level assurances. It must extend to how outputs are used, validated, and relied upon in practice. Finally, human rights provide the normative boundary conditions for all AI governance efforts. Ensuring that rights are protected requires not only technical safeguards but also institutional controls over how decisions are made and justified. Together, these priorities support a shift from governing systems in isolation to governing their use within accountable decision frameworks.

A critical cross-cutting issue not yet fully captured is the governance of human reliance on AI outputs in consequential decision-making. Existing frameworks emphasize system design, risk management, and deployment conditions. However, they do not sufficiently address when it is appropriate for individuals or institutions to rely on AI-generated outputs, under what verification conditions, and with what level of authority. This gap creates ambiguity in accountability. Even when systems meet technical and regulatory standards, inappropriate or unverified reliance on outputs can lead to significant institutional and societal harm. An emerging priority, therefore, is the development of operational decision thresholds that define different levels of reliance. For example, distinguishing between informational use, conditional reliance requiring verification, and contexts where reliance may be authorized under defined oversight structures. Addressing this issue would strengthen the connection between existing governance frameworks and real-world decision practices. It would also enhance auditability, clarify responsibility, and reduce systemic risk arising from overreliance or misuse of AI outputs. In this sense, the next phase of AI governance may require not only regulating systems, but also governing the conditions under which their outputs are relied upon.

In the Asia-Pacific region, and particularly within institutional sectors such as education and accreditation, current AI governance gaps are increasingly visible at the point where AI outputs are incorporated into real-world decisions. The most significant challenge is the disconnect between system-level governance and decision-level use. While frameworks have advanced in areas such as risk classification, model evaluation, and transparency, institutions lack clear operational guidance on when and how AI outputs can be relied upon in consequential contexts. This creates inconsistency in practice, where similar AI-assisted decisions may be subject to very different levels of verification, oversight, and accountability. As a result, institutional risk is no longer primarily technical—it emerges from uneven human reliance on AI outputs. In sectors such as education, this affects grading, placement, disciplinary decisions, and resource allocation, where insufficient verification or unclear authority structures can undermine both trust and fairness. At the same time, this gap presents a significant opportunity. Institutions in the region are well-positioned to serve as implementation environments for operationalizing AI governance. By defining clear verification thresholds, documentation standards, and escalation mechanisms, organizations can translate high-level principles into consistent decision practices. Additionally, there is an opportunity for the Asia-Pacific region to contribute to global interoperability efforts by demonstrating how governance frameworks can be aligned with real institutional workflows. This would reduce fragmentation, strengthen accountability, and enable more scalable and trustworthy adoption of AI. Addressing these gaps will be critical to ensuring that advances in AI governance translate into reliable, equitable, and auditable outcomes in practice.

The AI Dialogue can play a critical role in advancing international cooperation by functioning as a bridge between high-level governance principles and their operational implementation within institutions. At present, global AI governance efforts are characterized by strong normative alignment but limited convergence in how these principles are applied in practice. The Dialogue can address this by facilitating structured coordination around operational concepts—particularly how accountability, oversight, and verification are exercised when AI informs consequential decisions. One key role is to enable convergence around shared reference points that define responsible use across jurisdictions. This includes clarifying how institutions determine appropriate levels of human oversight, what constitutes sufficient verification, and how responsibility is documented and enforced. Without such alignment, interoperability across governance regimes will remain limited. Additionally, the Dialogue can serve as a platform for translating diverse national and sectoral experiences into practical guidance. By aggregating implementation insights from governments, International Organisations, and institutional actors, it can accelerate the development of governance approaches that are both context-sensitive and globally coherent. Finally, the Dialogue can support the development of voluntary coordination mechanisms—such as reporting structures, peer learning processes, or pilot initiatives—that test and refine governance practices in real-world settings. In this way, the Dialogue can move international cooperation from principle-level alignment toward actionable, interoperable governance that reflects how AI is actually used in decision-making contexts.

The AI Dialogue should build upon and connect existing international frameworks that have already established foundational principles and emerging implementation mechanisms. Key initiatives include the Organisation for Economic Co-operation and Development AI Principles and reporting efforts under the G7 Hiroshima AI Process, the United Nations Educational, Scientific and Cultural Organization Recommendation on the Ethics of Artificial Intelligence, the National Institute of Standards and Technology AI Risk Management Framework, and evolving international standards such as International Organisation for Standardization/IEC 42001. These frameworks provide strong normative and technical foundations but are often implemented in parallel, creating fragmentation at the institutional level. The added value of the AI Dialogue lies not in introducing new principles, but in enabling interoperability across these existing efforts. Specifically, the Dialogue can facilitate mapping across frameworks to identify alignment points, gaps, and areas of duplication, helping institutions navigate multiple governance expectations more efficiently. It can also promote shared operational interpretations of key concepts such as accountability, oversight, and risk thresholds, which are currently defined inconsistently across regimes. Furthermore, the Dialogue can elevate implementation-based insights by incorporating perspectives from institutions that operationalize these frameworks in practice. This would ensure that global governance efforts are informed not only by policy design but also by real-world constraints and decision-making dynamics. By acting as a coordination layer rather than a competing framework, the AI Dialogue can strengthen coherence, reduce fragmentation, and accelerate the practical adoption of trustworthy AI governance worldwide.

Different stakeholders should contribute not only perspectives, but also implementation experience tied to how AI governance operates in practice. Governments can provide regulatory direction and policy alignment. International Organisations can support coordination and interoperability across frameworks. The private sector can contribute technical insights and deployment realities. Civil society can ensure that societal and rights-based considerations are reflected. Critically, institutional actors—such as schools, hospitals, financial institutions, and accreditation bodies—should contribute operational perspectives on how AI is actually used in consequential decisions. To enable meaningful contributions, the Dialogue should be structured around three complementary layers. First, thematic plenaries to align on key governance priorities and emerging risks. Second, implementation-focused working sessions where stakeholders present concrete use cases, including how accountability, oversight, and verification are operationalized in practice. Third, synthesis sessions that translate discussions into structured outputs, such as reference frameworks, implementation guidance, or areas for continued collaboration. This structure would ensure that the Dialogue moves beyond general discussion toward actionable outcomes grounded in real-world application.

Global AI governance discussions often underrepresent institutional decision-makers who are directly responsible for applying AI in consequential contexts. While policymakers, technical experts, and industry leaders are well represented, there is comparatively limited input from those who must operationalize governance within real organizational environments. This includes school leaders, clinicians, judges, frontline public administrators, and accreditation or oversight bodies. These actors are responsible for determining how AI outputs are interpreted, validated, and acted upon, yet their perspectives are not systematically incorporated into governance design. In addition, there is a need to strengthen representation from regions and institutions with constrained resources, where governance must be implemented under practical limitations. Without these perspectives, frameworks risk being overly abstract or difficult to operationalize across diverse contexts. To address this gap, the Dialogue should actively include institutional practitioners through targeted invitations, structured case submissions, and dedicated sessions focused on implementation challenges. Mechanisms such as practitioner panels, field-based pilots, and documented case studies can ensure that governance discussions are informed by real decision-making conditions. Including these perspectives would strengthen the relevance, feasibility, and legitimacy of global AI governance efforts.

To foster meaningful engagement, the AI Dialogue should incorporate formats that prioritize interaction, practical insight, and iterative learning rather than solely formal presentations. One effective approach is the use of structured case-based discussions, where participants examine real or simulated scenarios involving AI-assisted decision-making. This allows stakeholders to explore how governance principles apply under specific conditions, including questions of accountability, verification, and oversight. Another valuable format is small-group "implementation labs," where diverse stakeholders collaboratively work through governance challenges and develop practical solutions. These sessions can generate concrete outputs such as decision protocols, documentation standards, or risk thresholds. In addition, the Dialogue could incorporate peer exchange mechanisms, where institutions share lessons learned from implementing AI governance in different sectors and regions. This would enable cross-context learning and accelerate the diffusion of effective practices. Finally, incorporating iterative feedback loops—such as pre-submitted inputs, live synthesis sessions, and post-dialogue follow-ups—would help ensure continuity and sustained engagement beyond the event itself. These formats would shift the Dialogue from a forum for discussion to a platform for co-developing actionable and context-relevant governance approaches.

Several existing policies and practices offer strong foundations for effective AI governance, particularly when they are translated into operational mechanisms within institutions. The Organisation for Economic Co-operation and Development AI Principles have been influential in establishing a shared normative baseline, particularly around transparency, accountability, and human-centered values. Similarly, the United Nations Educational, Scientific and Cultural Organization Recommendation on the Ethics of Artificial Intelligence provides a comprehensive framework linking governance to human rights and societal impact. From an operational perspective, the National Institute of Standards and Technology AI Risk Management Framework offers a structured approach to identifying, assessing, and managing AI-related risks across the system lifecycle. Emerging standards such as International Organisation for Standardization/IEC 42001 further contribute by formalizing management system requirements for AI governance. A key good practice across these efforts is the integration of governance into existing institutional processes rather than treating it as a standalone compliance function. This includes embedding documentation requirements, audit mechanisms, and review protocols into routine decision workflows. However, a critical area for further development is the translation of these frameworks into decision-level practices. Effective approaches are beginning to emerge where institutions define clear conditions for how AI outputs are used, including verification requirements, human oversight structures, and escalation pathways when outputs are uncertain or contested. These practices demonstrate that effective AI governance depends not only on robust frameworks, but on their consistent application within real-world decision-making environments.