Middle States Association (Commissioner) | OECD HAIP Task Force Contributor

A successful first Global Dialogue on AI Governance should move beyond high-level principles and establish a clear pathway toward operational governance—specifically, how institutions make and justify decisions using AI systems. Three outcomes would signal success. First, conceptual clarity on where risk materializes. Current frameworks emphasize model development and evaluation, yet institutional risk emerges at the moment a decision is taken based on AI-assisted outputs. The Dialogue should explicitly recognize this "point of action" as a core governance boundary. Second, convergence toward interoperable decision-governance standards. Rather than duplicating existing frameworks, the Dialogue should advance alignment across major regimes (e.g., EU AI Act, OECD AI Principles, NIST AI RMF) by defining common expectations for accountability, verification, and human oversight at the decision level. Third, a commitment to implementation-oriented outputs. This includes guidance that is auditable, enforceable, and usable by deploying institutions—not only developers. Practical artifacts could include model documentation requirements linked to decision use, minimum verification thresholds for high-stakes applications, and traceability expectations for AI-informed decisions. Ultimately, success will be measured not by consensus alone, but by whether institutions can operationalize governance in real-world contexts. The Dialogue should therefore prioritize decision integrity: ensuring that when AI informs consequential actions, those actions remain justified, accountable, and subject to meaningful oversight.

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

These priorities reflect the need to move from principles to enforceable governance at the point where AI systems influence real-world decisions. Safe, secure and trustworthy AI is foundational, but trust must extend beyond system performance to the conditions under which outputs are relied upon in consequential actions. This requires clarity on when AI-assisted analysis is sufficient, when independent verification is required, and when reliance must be restricted. Interoperability of governance approaches is essential given the rapid emergence of parallel frameworks across jurisdictions. Without alignment, institutions face fragmentation in compliance expectations. Interoperability should therefore focus on harmonizing accountability structures, verification thresholds, and documentation requirements across regimes. Transparency, accountability, and human oversight are most effective when operationalized at the decision level. Rather than general oversight requirements, governance should define how decisions informed by AI are documented, reviewed, and justified-particularly in high-stakes contexts. Protection and promotion of human rights remains a central objective, but its practical realization depends on enforceable governance mechanisms. Rights are most at risk not at the point of model design alone, but when institutions act on outputs without sufficient safeguards, traceability, or recourse. Together, these priorities support a shift toward decision-centered governance, ensuring that AI-enabled actions remain accountable, verifiable, and aligned with both legal and ethical standards.

A critical gap across existing themes is the absence of a defined framework for institutional reliance on AI-assisted outputs-specifically, when it is appropriate for organizations to act on such outputs in consequential decisions. Current governance efforts largely focus on system-level properties (e.g., accuracy, robustness, bias), yet these do not directly determine whether a decision informed by AI is justified. The central governance challenge lies in the transition from analysis to action: how institutions authorize, verify, and take accountability for decisions involving AI. This gap can be addressed through the introduction of decision-governance thresholds. For example, AI outputs may serve as informational inputs to human judgment, require independent verification prior to use, or support bounded operational decisions under defined authority and oversight. These distinctions are not consistently reflected in current frameworks but are essential for managing real-world risk. A related cross-cutting issue is the need to align accountability with decision authority. As AI systems become more integrated into workflows, responsibility can become diffused across developers, deployers, and decision-makers. Governance must therefore clarify who holds authority at the moment of action and under what conditions that authority is exercised. Addressing these issues would strengthen the Dialogue's ability to move from abstract principles to actionable governance, ensuring that institutional decisions involving AI remain justified, auditable, and subject to meaningful oversight.

Across education and regulated institutional environments, the most significant governance gap is not the availability of AI systems, but the absence of clear standards governing when institutions are justified in acting on AI-assisted outputs. In practice, AI tools are increasingly used to inform high-stakes decisions—such as student progression, disciplinary actions, resource allocation, and institutional evaluations. However, existing governance frameworks primarily address system-level properties (e.g., accuracy, bias, robustness) rather than the conditions under which outputs may be relied upon in consequential decisions. This creates a critical ambiguity at the point of action, where accountability, verification, and authority are often insufficiently defined. The result is uneven practice across institutions. Some over-restrict AI use due to liability concerns, limiting potential benefits, while others rely on outputs without adequate verification, exposing stakeholders to procedural and ethical risks. This inconsistency is further compounded in cross-border contexts, where differing regulatory expectations create fragmentation and uncertainty for institutions operating internationally. At the same time, there is a significant opportunity to strengthen governance by shifting toward decision-centered approaches. By defining operational thresholds for AI use—distinguishing between informational support, conditional reliance requiring verification, and bounded authorization for action—institutions can align innovation with accountability. Such an approach would also support interoperability across existing frameworks by linking system-level evaluation to decision-level governance. For sectors like education, where decisions carry long-term human impact and formal accountability, addressing this gap is essential. Advancing governance at the level of institutional decision-making will enable safer adoption of AI while preserving trust, fairness, and procedural integrity.

The AI Dialogue can play a critical role by shifting international cooperation from principle alignment to operational governance—specifically, by defining how institutions across jurisdictions make and justify decisions involving AI. Most existing efforts have successfully established high-level principles, yet cooperation remains fragmented at the point of implementation. The Dialogue can address this by serving as a coordination layer that translates global norms into interoperable decision-governance expectations. First, the Dialogue can establish a shared understanding of where governance responsibility materializes: at the moment institutions act on AI-assisted outputs. Recognizing this "point of action" as a common governance boundary would enable more consistent approaches to accountability, verification, and oversight across jurisdictions. Second, it can facilitate convergence by identifying common decision-level requirements that can be mapped across existing frameworks. This includes expectations for when AI outputs may inform decisions, when independent verification is required, and how decision processes should be documented and audited. Third, the Dialogue can provide a neutral platform for multistakeholder input focused on deployer-level realities, ensuring that governance frameworks are not limited to system developers but reflect the conditions under which institutions operate. In this way, the AI Dialogue can add value not by duplicating existing initiatives, but by enabling interoperability at the level of institutional decision-making—where governance must ultimately be applied and enforced.

The AI Dialogue should build upon and connect existing frameworks that have already established foundational principles and risk management approaches, including the OECD AI Principles, the G7 Hiroshima AI Process, the EU AI Act, the NIST AI Risk Management Framework, and emerging ISO/IEC standards such as ISO/IEC 42001. These initiatives collectively provide a strong base across ethics, risk classification, system evaluation, and governance processes. However, they remain partially fragmented in their operational application—particularly in how institutions translate these principles into real-world decisions. The added value of the AI Dialogue lies in its ability to act as a unifying coordination layer across these efforts. Specifically, it can advance interoperability by aligning how governance is implemented at the decision level, rather than only at the system level. This includes bridging gaps between developer-focused standards and deployer responsibilities, clarifying expectations for institutional accountability, and establishing shared approaches to verification, documentation, and oversight when AI informs consequential actions. In addition, the Dialogue can provide a global forum for integrating perspectives from sectors that operate under formal accountability—such as education, healthcare, and public administration—where governance must be auditable and enforceable in practice. By connecting existing frameworks through a decision-centered lens, the AI Dialogue can move global governance from parallel systems toward coordinated implementation, ensuring that principles translate into consistent, actionable standards across jurisdictions.

Different stakeholders should contribute based on where they exercise authority within the AI lifecycle—developers, deployers, regulators, and affected communities—rather than through generalized participation alone. The Dialogue should therefore be structured to reflect these distinct roles. A useful format would combine three layers. First, thematic plenaries to align on shared principles and emerging risks. These should remain concise and outcome-oriented, focused on identifying areas requiring coordination rather than restating existing frameworks. Second, role-based working groups organized around decision responsibility. For example, developers, institutional deployers, regulators, and civil society actors should each contribute perspectives grounded in their operational realities. This enables more precise identification of where governance obligations arise and how they are implemented in practice. Third, cross-stakeholder synthesis sessions focused on translating inputs into actionable governance elements. These sessions should prioritize convergence on implementable expectations, such as verification requirements, documentation standards, and accountability structures when AI informs consequential decisions. To ensure effectiveness, contributions should be structured through targeted prompts that require participants to specify not only risks and principles, but also conditions for use, oversight mechanisms, and decision authority. In this way, the AI Dialogue can move beyond broad inclusivity toward structured participation that reflects how governance is actually exercised, ensuring outputs are operational, interoperable, and applicable across jurisdictions.

A key underrepresented perspective in global AI governance discussions is that of institutional decision-makers operating in regulated environments—those responsible for acting on AI-assisted outputs under conditions of formal accountability. While significant attention has been given to governments, developers, and civil society, less emphasis has been placed on deployers within institutions such as schools, hospitals, and public agencies. These actors are where governance is ultimately realized in practice, yet their constraints—legal liability, evidentiary requirements, and procedural oversight—are often insufficiently reflected in global frameworks. Additionally, cross-border institutions operating across multiple regulatory regimes remain underrepresented. These organizations must navigate fragmented governance expectations while maintaining consistent decision standards, making their experience critical for advancing interoperability. Perspectives from frontline practitioners are also essential. Individuals responsible for implementing decisions—teachers, administrators, case workers—encounter the practical limitations of AI systems and governance requirements, yet their insights are rarely systematically integrated into policy discussions. To address these gaps, the Dialogue should actively include deployer-level actors through targeted invitations, structured case submissions, and sector-specific working groups. Mechanisms such as anonymized decision case studies and implementation reports could provide grounded evidence of how governance functions in real contexts. Including these perspectives would strengthen the Dialogue's ability to produce governance approaches that are not only principled, but also enforceable, context-aware, and aligned with real-world accountability structures.

To foster meaningful engagement, the AI Dialogue should move beyond traditional panel discussions and incorporate formats that reflect how governance decisions are made in practice. One effective approach would be structured "decision scenario simulations," where participants are presented with real-world cases involving AI-assisted decision-making across sectors. Stakeholders would be required to determine whether and how action should be taken, specifying conditions for reliance, verification requirements, and accountability. This format would surface differences in governance assumptions and enable convergence on practical standards. A second format would involve cross-framework mapping exercises, where participants align governance expectations across existing initiatives (e.g., OECD, EU AI Act, NIST). This would support interoperability by identifying common elements at the level of implementation, rather than remaining at the level of principles. Third, the Dialogue could incorporate "evidence-based submissions" in which participants present short, structured case inputs documenting how AI is used in consequential decisions, including challenges, safeguards, and outcomes. These could be synthesized into a shared repository to inform ongoing policy development. Finally, small, facilitated working sessions focused on producing draft governance elements—such as decision thresholds, documentation standards, or oversight protocols—would ensure that engagement leads to tangible outputs. These formats prioritize applied reasoning, comparability, and implementation, enabling the Dialogue to generate insights that are directly transferable to institutional governance contexts.

Effective AI governance is emerging through a combination of established frameworks and implementation-level practices that translate principles into operational controls. At the policy level, several initiatives provide strong foundations. The OECD AI Principles and the G7 Hiroshima AI Process establish globally recognized expectations for trustworthy AI. The EU AI Act advances a risk-based regulatory model, while the NIST AI Risk Management Framework offers a structured approach to identifying, assessing, and mitigating AI-related risks. Emerging standards such as ISO/IEC 42001 further support the institutionalization of AI management systems. Together, these frameworks define essential elements of transparency, accountability, and risk governance. However, effective governance depends on how these principles are operationalized within institutions. One emerging practice is the introduction of decision-level governance controls that define how AI-assisted outputs may be used in consequential actions. This includes distinguishing between uses where AI informs human judgment, where outputs require independent verification, and where bounded reliance may be authorized under defined conditions. Additional good practices include formal documentation of AI-assisted decisions, traceability mechanisms linking outputs to actions, and structured review processes to ensure procedural integrity. In regulated environments, governance is strengthened when decision-making processes are auditable and aligned with existing accountability frameworks. Platforms that support governance implementation-such as reporting mechanisms aligned with international frameworks and institutional-level transparency disclosures-also contribute to consistency and comparability across jurisdictions. These approaches demonstrate that effective AI governance requires both high-level coordination and practical implementation mechanisms. Bridging this gap-particularly at the level of institutional decision-making-will be critical to ensuring that AI systems are deployed responsibly, consistently, and in a manner that upholds accountability and trust.