HQX

A successful first Global Dialogue on AI Governance should deliver more than alignment of perspectives; it should establish the foundations for verifiable trust in AI systems at a global scale. First, success would mean moving beyond high-level principles toward operational clarity. This includes identifying concrete mechanisms through which governance commitments can be implemented, monitored, and independently verified across jurisdictions and sectors. Second, the Dialogue should produce a shared baseline for interoperability. With multiple regulatory frameworks emerging (e.g., EU AI Act, national strategies, sectoral rules), aligning on minimum standards for safety, accountability, and transparency is critical to avoid fragmentation and regulatory arbitrage, while enabling innovation. Third, it should meaningfully address the AI divide, not only in terms of access to technology, but also access to governance capacity. This includes empowering emerging economies with tools, infrastructure, and institutional frameworks to participate in shaping, not just adopting, AI systems. Fourth, a successful outcome would be the recognition that trust in AI cannot rely solely on explainability or policy commitments. It requires the ability to verify how systems are built, deployed, and used. Introducing the concept of verifiable governance, including traceability, auditability, and, where appropriate, cryptographic assurance layers, would mark an important step forward. Finally, the Dialogue should result in a clear forward pathway, such as a roadmap or working groups, ensuring continuity beyond Geneva and translating discussion into sustained, multi-stakeholder action. In essence, success lies in shifting from dialogue to demonstrable accountability.

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

My selection reflects a focus on advancing AI governance from principles to verifiable, operational frameworks. Safe, secure and trustworthy AI is foundational, particularly as AI systems are increasingly embedded in high-impact domains such as finance, healthcare, and public administration. Trust must be engineered, not assumed. Transparency, accountability, and human oversight are central to ensuring that AI systems can be meaningfully scrutinized. However, transparency alone is insufficient. There is a growing need to move toward verifiable accountability, where decisions can be traced, audited, and validated against defined governance controls. Interoperability of governance approaches is critical in a fragmented global landscape. With multiple regulatory regimes emerging, alignment on baseline standards and mechanisms is necessary to enable cross-border collaboration while reducing regulatory arbitrage and implementation friction. Protection and promotion of human rights provides the normative anchor for all governance efforts. As AI systems increasingly influence rights-sensitive decisions, governance frameworks must ensure that fundamental rights are not only protected in principle but upheld in practice through enforceable and measurable safeguards. Across these priorities, my work focuses on the intersection of governance, technology, and trust infrastructure, contributing to approaches that combine policy, technical design, and verification mechanisms to strengthen accountability in AI systems.

Yes. While the listed themes capture key dimensions of AI governance, several cross-cutting and emerging issues remain insufficiently articulated. First, the need for verifiability as a distinct governance layer. Current frameworks emphasize transparency and accountability, but often rely on declarations, documentation, or explainability. In high-impact contexts, this is not enough. There is a growing need for mechanisms that allow independent verification of how AI systems are built, trained, and deployed, including traceability of data, model lineage, and decision pathways. This introduces a shift from trust by disclosure to trust by evidence. Second, dependency and concentration risk in AI infrastructure. As organizations increasingly rely on a small number of model providers and platforms, systemic vulnerabilities emerge. Sudden loss of access, opaque enforcement actions, or infrastructure failures can disrupt critical operations. Governance discussions should therefore address resilience, redundancy, and controllability as core requirements. Third, the rise of machine-to-machine ecosystems and agentic AI. As autonomous and semi-autonomous systems begin to interact with each other, governance must evolve beyond human-in-the-loop assumptions. This raises new questions around responsibility, auditability, and control in distributed decision environments. Fourth, the gap between governance design and implementation capacity. Many institutions, particularly in emerging economies, lack the technical and organizational infrastructure to operationalize governance frameworks. Bridging this gap requires not only policy alignment but deployable tools and institutional capability-building. These issues cut across all thematic areas and point toward a broader shift: AI governance must evolve from principles and compliance toward resilient, verifiable, and system-level accountability.

Governance gaps in AI are increasingly visible across sectors where systems influence high-stakes decisions, particularly in finance, digital services, and public-facing platforms. One of the most significant challenges is the gap between policy commitments and operational enforcement. While regulatory frameworks are advancing, organizations often lack the mechanisms to demonstrate that AI systems are functioning within approved parameters. This creates exposure to legal, reputational, and systemic risks, especially as scrutiny increases. A second challenge is fragmentation across jurisdictions. With different regulatory approaches emerging globally, organizations operating across borders face complexity in aligning governance, compliance, and technical implementation. This not only increases costs but can also slow down responsible innovation. A third issue is infrastructure dependency risk. Many institutions rely on external AI providers and centralized platforms, creating vulnerabilities related to access, continuity, and control. This has direct implications for operational resilience and strategic autonomy. At the same time, these gaps present important opportunities. There is a growing opportunity to develop governance as infrastructure, embedding accountability directly into system design through traceability, auditability, and verification mechanisms. This enables organizations to move from reactive compliance to proactive trust-building. Additionally, the current landscape creates space for interoperable governance models, where shared standards and technical approaches can reduce fragmentation while maintaining flexibility across regions. Finally, there is a strong opportunity to position governance not as a constraint, but as a competitive advantage, particularly for organizations and jurisdictions that can demonstrate reliable, accountable, and trustworthy AI systems in practice. Overall, the shift underway is from abstract governance to enforceable, system-level accountability.

The AI Dialogue can play a critical role by acting as a bridge between principles and implementation, enabling international cooperation to move from alignment of intent to coordination of action. First, it can help establish a shared operational baseline. While many jurisdictions have articulated high-level principles, there is less convergence on how these are implemented in practice. The Dialogue can facilitate agreement on common governance components such as risk classification approaches, audit requirements, and minimum safeguards, supporting interoperability across regulatory systems. Second, it can serve as a platform to advance interoperable governance mechanisms, reducing fragmentation. By bringing together governments, industry, academia, and civil society, the Dialogue can identify areas where alignment is feasible without requiring uniform regulation, enabling cross-border collaboration while respecting national contexts. Third, the Dialogue can elevate the importance of verifiable accountability. International cooperation will increasingly depend not only on shared rules, but on the ability to demonstrate compliance in a consistent and credible way. Promoting approaches that integrate traceability, auditability, and independent verification would strengthen trust between jurisdictions and institutions. Fourth, it can address asymmetries in capacity and access. By supporting knowledge exchange, technical collaboration, and practical toolkits, the Dialogue can help ensure that emerging economies are not only recipients of AI systems but active participants in shaping governance frameworks. Finally, the Dialogue can create continuity beyond convening, for example through working groups, pilot initiatives, or reference models that translate discussions into sustained cooperation. In this sense, its role is not only to convene stakeholders, but to catalyze a shift toward coordinated, credible, and enforceable global AI governance.

The AI Dialogue should build on existing global, regional, and multi-stakeholder initiatives that are already shaping the governance landscape, while addressing the gaps between them. Key reference points include the OECD AI Principles and the G20 AI frameworks, which provide widely endorsed normative foundations; UNESCO's Recommendation on the Ethics of AI, which integrates human rights and societal considerations; and the Global Partnership on AI (GPAI), which advances applied research and policy collaboration. At the regulatory level, frameworks such as the EU AI Act and emerging national strategies offer concrete approaches to risk classification and compliance. In parallel, industry-led and multi-stakeholder initiatives such as the Frontier Model Forum, the AI Safety Institutes network, and technical standard-setting bodies (e.g., ISO/IEC, IEEE) are contributing to safety practices, evaluation methods, and technical standards. While these efforts are valuable, they remain fragmented across domains, geographies, and levels of maturity. The added value of the AI Dialogue lies in its ability to act as a convergence layer. It can connect normative frameworks with regulatory approaches and technical standards, helping translate principles into interoperable practices. Importantly, the Dialogue can also advance alignment on operational mechanisms, such as auditability, certification approaches, and evidence-based compliance, which are not yet consistently defined across initiatives. In addition, it can provide a neutral space to address cross-cutting risks, including systemic dependencies and governance gaps in rapidly evolving areas such as generative and agentic AI. Finally, by bringing together diverse stakeholders under a UN-convened process, the Dialogue can strengthen global legitimacy and inclusiveness, ensuring that governance approaches reflect not only leading economies but a broader set of perspectives. In this sense, its role is to connect, align, and operationalize what currently exists in parallel.

Effective contribution to the AI Dialogue requires moving beyond representation toward structured, outcome-oriented participation across stakeholder groups. Governments should provide regulatory perspectives, share implementation experiences, and identify areas where alignment or mutual recognition is feasible. Industry should contribute practical insights on system design, deployment challenges, and risk management, including lessons from real-world use. Academia and technical experts should support with independent evaluation methods, metrics, and emerging research, particularly in safety, robustness, and auditability. Civil society plays a critical role in grounding discussions in societal impact, human rights, and accountability from an end-user perspective. To maximize effectiveness, the Dialogue should adopt a multi-layered structure: First, combine high-level plenary sessions with focused technical working groups. Plenaries can align on priorities, while smaller groups address specific topics such as risk frameworks, audit mechanisms, or interoperability. Second, organize discussions around concrete use cases and sector-specific scenarios (e.g., finance, healthcare, public services). This helps translate abstract principles into operational realities. Third, include output-oriented sessions designed to produce tangible deliverables, such as draft guidelines, reference models, or shared taxonomies, rather than open-ended exchanges. Fourth, enable cross-regional participation across time zones and ensure continuity through virtual follow-ups, maintaining momentum beyond the main convening. Finally, incorporate feedback and validation loops, allowing stakeholders to test and refine proposed approaches in practice. Overall, the Dialogue should be structured not only as a forum for discussion, but as a platform for co-creating implementable, interoperable, and verifiable governance approaches

Several voices remain underrepresented in global AI governance, despite being directly affected by its outcomes. First, practitioners responsible for implementation, including compliance officers, risk managers, auditors, and system integrators, are often missing. These actors translate policy into operational reality and can identify where governance frameworks fail in practice. Including them through technical working groups and implementation-focused sessions would strengthen feasibility. Second, communities in the Global South remain underrepresented not only in participation but in agenda-setting. This includes policymakers, local innovators, and institutions that face distinct constraints in infrastructure, data access, and regulatory capacity. Their inclusion requires more than invitations; it calls for funded participation, regional consultations, and mechanisms to incorporate their priorities into outcomes. Third, small and medium-sized enterprises (SMEs) and startups are often absent from discussions dominated by large technology companies. Yet they face disproportionate compliance burdens and have limited resources to navigate fragmented governance landscapes. Dedicated SME tracks or advisory groups could ensure their perspectives inform proportional and scalable approaches. Fourth, affected communities and end-users, particularly in high-impact domains such as social services, labor, and financial access, are rarely directly represented. Their lived experience is critical to understanding real-world implications of AI systems. Structured engagement through civil society organizations and participatory mechanisms would help address this gap. Finally, there is a lack of interdisciplinary voices, particularly from law, behavioral science, and ethics, integrated with technical expertise. Bridging these perspectives is essential for holistic governance. A more inclusive AI Dialogue requires intentional design: resourcing participation, decentralizing consultations, and embedding diverse perspectives into decision-shaping processes, not just consultation stages.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional panels toward interactive, outcome-driven formats that reflect the complexity of AI governance. First, use-case simulation labs can be highly effective. Participants work through real-world scenarios (e.g., AI in credit scoring or public services), mapping risks, governance controls, and accountability gaps. This grounds discussions in operational reality and reveals where frameworks succeed or fail. Second, multi-stakeholder "design sprints" can bring together governments, industry, academia, and civil society to co-develop specific outputs within a defined timeframe, such as a draft audit model, risk taxonomy, or interoperability guideline. This shifts engagement from dialogue to co-creation. Third, red-teaming and stress-testing exercises can be introduced to challenge assumptions. By simulating failures, misuse, or regulatory edge cases, participants can better understand systemic vulnerabilities and strengthen governance approaches. Fourth, structured debate formats (e.g., Oxford-style debates) on contested issues—such as open vs. closed models or national vs. global regulation—can surface trade-offs more clearly than consensus-driven discussions. Fifth, interactive demonstration sessions can allow technical experts to showcase tools for traceability, auditing, or verification, helping bridge the gap between policy and implementation. Finally, the Dialogue should include continuous digital collaboration spaces before and after the event, enabling asynchronous input, iteration of ideas, and broader participation across time zones. These formats collectively support a shift from passive exchange to active problem-solving, helping generate practical, testable, and globally relevant governance outcomes.

Several practices are beginning to move AI governance from theory into something more tangible and effective. One important example is the EU AI Act, which introduces a risk-based approach. Its strength is not only in defining categories, but in linking them to concrete obligations. It gives organizations a clearer path from principle to implementation, even if challenges remain in execution. Another meaningful development is the emergence of AI Safety Institutes and model evaluation frameworks. These initiatives focus on testing systems before and after deployment, creating a culture where performance, safety, and limitations are actively examined rather than assumed. On the organizational side, we are seeing promising practices in internal AI governance structures. Some companies are establishing cross-functional AI oversight committees, combining legal, technical, and risk perspectives. This helps ensure that decisions are not made in silos and that accountability is shared. There are also practical advances in traceability and documentation, such as model cards and data sheets. While still evolving, they represent an important step toward making AI systems more understandable and auditable. At a more technical level, approaches that integrate auditability and verification mechanisms are gaining attention. This includes logging systems, model lineage tracking, and, in some cases, cryptographic methods to ensure the integrity of records. These approaches help move governance from statements to evidence. Finally, multi-stakeholder initiatives such as OECD, UNESCO, and GPAI play an important role in aligning values and creating shared language across regions. Taken together, these efforts show a clear direction: effective AI governance emerges when policy, technical design, and organizational practices are developed together, not in isolation.