H21 Visions

A successful first Global Dialogue on AI Governance should move beyond principles toward operational convergence. First, it should deliver a shared baseline for verifiable AI governance—not uniform regulation, but interoperable standards that allow systems to demonstrate compliance across jurisdictions. Trust cannot scale through declarations alone; it requires testable, auditable mechanisms. Second, the Dialogue should establish a common language for accountability, including measurable indicators for system integrity, risk thresholds, and oversight responsibilities. This enables comparability between governance regimes without forcing regulatory homogeneity. Third, it should launch pilot frameworks for cross-border verification, where governments, companies, and technical bodies test how AI systems can produce proof of compliance (e.g., through cryptographic or audit-based methods). This shifts governance from static compliance to continuous assurance. Fourth, success requires embedding governance interoperability as a strategic priority, ensuring that fragmented regulatory approaches do not create systemic risk or exclusion—especially for emerging economies. Finally, the Dialogue should result in a clear implementation roadmap: timelines, responsible actors, and mechanisms for iterative coordination. Without execution pathways, alignment efforts risk remaining symbolic. In short, success is achieved when the Dialogue transforms AI governance from a fragmented, principle-based discussion into a globally interoperable system of verifiable trust.

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

The selected priorities-safe and trustworthy AI, interoperability of governance approaches, transparency and accountability, and the broader societal implications of AI-are interconnected dimensions of a single challenge: how to make trust in AI systems scalable across jurisdictions. Safety without verifiability remains aspirational. Transparency without comparability remains fragmented. And governance without interoperability risks reinforcing geopolitical and technological asymmetries. Interoperability is therefore central: it allows different regulatory frameworks to coexist while still enabling systems to demonstrate compliance in a consistent, testable way. This is critical in a global AI ecosystem where systems operate across borders but are governed locally. Transparency and accountability must evolve from static disclosures toward continuous, auditable signals of system integrity, enabling regulators, organizations, and users to assess risk dynamically. At the same time, the social, economic, and cultural implications of AI cannot be treated as secondary. Governance decisions define not only technical outcomes, but also who benefits, who bears risk, and how power is distributed in digital systems. Together, these priorities reflect a shift from governance as policy alignment to governance as system design-where trust is not assumed, but engineered, measured, and maintained.

A critical cross-cutting issue is the need to move from principle-based governance to verifiable governance infrastructure. Current frameworks emphasize ethics, safety, and accountability, but lack mechanisms to prove compliance in real time and across jurisdictions. This creates a structural gap between regulatory intent and operational reality. A key emerging priority is the development of machine-verifiable compliance systems, where AI models and services can generate auditable evidence of adherence to regulatory requirements without exposing sensitive data. This includes cryptographic approaches, standardized integrity metrics, and continuous monitoring systems. Another gap is governance at the infrastructure layer-including compute concentration, data pipelines, and model supply chains. Without visibility into these layers, oversight remains incomplete. Additionally, temporal governance is underdeveloped: AI systems evolve after deployment, yet regulatory frameworks are largely static. Mechanisms for continuous validation and lifecycle accountability are essential. Finally, there is a need to address asymmetries in governance capacity. Without tools that enable smaller states and organizations to verify and enforce standards, global AI governance risks becoming structurally unequal. Addressing these issues would shift AI governance from reactive oversight to proactive, system-level integrity engineering.

Across regions and sectors, the primary governance gap is not a lack of frameworks, but a lack of operational interoperability and verifiability. In Europe, regulatory leadership (e.g., risk-based AI frameworks) is creating important safeguards, but also exposing fragmentation: systems deployed across borders must navigate non-aligned compliance requirements without a common mechanism to demonstrate equivalence. This increases costs, slows innovation, and creates uncertainty—particularly for smaller actors. A second gap lies in static compliance models. Current approaches rely heavily on documentation and ex-ante classification, while AI systems evolve continuously post-deployment. This creates a mismatch between regulatory oversight and system behavior, limiting the ability to monitor risk in real time. Third, there is limited capacity for technical verification at scale. Many organizations can declare compliance, but few can produce auditable, machine-verifiable evidence of safety, integrity, or alignment with governance standards. However, these gaps also create significant opportunities. First, there is a clear opening to develop interoperable governance infrastructures, where different regulatory regimes can be connected through shared metrics, taxonomies, and verification layers rather than harmonized into a single model. Second, advances in cryptography, auditing systems, and secure computation enable a shift toward continuous assurance, where compliance becomes dynamic and testable rather than static and declarative. Third, regions that invest early in verifiable governance capabilities—including tools, standards, and institutional capacity—can position themselves as trusted hubs in the global AI ecosystem. Ultimately, the opportunity is to transition from fragmented oversight to scalable trust architectures, where governance is embedded into the technical and operational fabric of AI systems.

The AI Dialogue can play a pivotal role by shifting international cooperation from norm alignment to operational interoperability. Today, most cooperation efforts focus on shared principles, yet AI systems operate across jurisdictions with incompatible compliance requirements and no common verification layer. The Dialogue can address this gap by fostering the development of interoperable governance infrastructures, enabling different regulatory regimes to remain sovereign while still being technically compatible. Specifically, the Dialogue can: * Promote a shared framework for verifiable compliance, including common metrics, taxonomies, and audit mechanisms that allow AI systems to demonstrate adherence across jurisdictions. * Enable cross-border pilot programs, where governments and industry test real-world interoperability of governance approaches, moving from theoretical alignment to applied coordination. * Facilitate technical cooperation on assurance mechanisms, such as continuous monitoring, secure auditing, and privacy-preserving verification methods. * Support the creation of a common language for accountability, allowing regulators to interpret and compare system risks and governance outcomes consistently. Importantly, the Dialogue should not aim to standardize regulation globally, but to ensure that different systems can communicate, translate, and verify each other's requirements. In doing so, it can reduce fragmentation, lower compliance friction, and strengthen trust in cross-border AI deployment. Ultimately, its role is to enable a transition from fragmented governance to coordinated, testable trust systems at a global scale.

The AI Dialogue should build on existing initiatives while addressing their shared limitation: a gap between principles and operational verification. Relevant foundations include: * International normative frameworks (e.g., UNESCO AI Ethics Recommendation, OECD AI Principles) * Regulatory approaches (e.g., EU AI Act and other risk-based models) * Technical standards bodies (e.g., ISO/IEC, IEEE) * Multi-stakeholder initiatives (e.g., GPAI, Partnership on AI) These efforts have established important common ground, but remain largely non-interoperable at the implementation level. The added value of the AI Dialogue lies in acting as a coordination layer across these ecosystems, with three key contributions: 1. Interoperability bridging: translating high-level principles and regulatory requirements into shared technical standards and verification protocols that can operate across jurisdictions. 2. From compliance to assurance: advancing mechanisms for continuous, auditable validation of AI systems, rather than static, documentation-based compliance. 3. Cross-ecosystem integration: connecting policymakers, technical bodies, and industry actors to co-develop practical governance tools, including testing environments, shared metrics, and pilot infrastructures. Additionally, the Dialogue can help ensure that emerging economies are not excluded by enabling accessible verification tools and capacity-building mechanisms. In essence, its value is not to duplicate existing efforts, but to make them work together—transforming a fragmented landscape into a functioning global governance architecture.

Different stakeholders should contribute not only perspectives, but operational capabilities. * Governments: define policy objectives, risk thresholds, and public accountability frameworks. * Industry: provide real-world deployment insights and participate in pilot implementations of verifiable governance mechanisms. * Technical community: develop standards, metrics, and verification protocols. * Civil society and academia: assess societal impact, legitimacy, and long-term risks. To be effective, the Dialogue should be structured as a working system, not a conference. First, organize it around thematic working groups (e.g., interoperability, assurance mechanisms, accountability metrics), each tasked with producing concrete outputs (standards drafts, pilot designs, evaluation frameworks). Second, integrate policy–technical labs, where stakeholders co-develop and test solutions (e.g., cross-border compliance verification, continuous monitoring tools). Third, establish feedback loops between high-level discussions and technical implementation, ensuring that principles translate into deployable mechanisms. Fourth, include regional nodes to capture context-specific challenges while feeding into a global coordination layer. Finally, define clear deliverables and timelines, including pilot programs, shared taxonomies, and verification frameworks. The Dialogue should function as a coordination and production platform, enabling stakeholders to jointly build the infrastructure of interoperable AI governance.

Global AI governance discussions remain structurally imbalanced, with several critical voices underrepresented. These include: * Emerging and smaller economies, which often lack the capacity to shape or verify governance standards but are deeply affected by them. * Technical implementers outside major platforms, including SMEs and open-source communities, who face disproportionate compliance burdens. * Communities impacted by AI deployment, whose lived experience of system failures or biases is rarely translated into governance design. * Interdisciplinary perspectives, particularly from social sciences, humanities, and cultural sectors, which are essential to understanding systemic impact. Inclusion requires moving beyond participation toward capability enablement. First, provide access to shared governance tools, including open verification frameworks and technical resources that lower barriers to entry. Second, establish funded participation mechanisms to ensure sustained engagement from underrepresented regions and groups. Third, create structured input channels, where experiential knowledge (e.g., impact assessments, case studies) is systematically integrated into governance outputs. Fourth, support regional co-development hubs, enabling local actors to adapt and contribute to global standards. Inclusion should not be symbolic—it should ensure that diverse actors can actively shape and implement governance systems, not just respond to them.

To foster meaningful engagement, the Dialogue should prioritize interactive, outcome-oriented formats over traditional panels. Key approaches include: * Policy–technical simulation labs: stakeholders test governance scenarios (e.g., cross-border AI deployment, compliance verification) in controlled environments, identifying gaps and solutions in real time. * Interoperability "sandboxes": jurisdictions and organizations experiment with how different regulatory frameworks can be translated and validated across systems. * Live audit demonstrations: showcasing how AI systems can generate real-time, verifiable evidence of compliance, making governance tangible and testable. * Scenario-based governance exercises: exploring future risks (e.g., systemic failures, cross-border incidents) to stress-test coordination mechanisms. * Collaborative drafting sessions: producing shared outputs (standards, metrics, pilot frameworks) during the Dialogue itself. Additionally, the Dialogue should incorporate continuous digital participation layers, allowing contributions before, during, and after in-person sessions, ensuring broader and sustained engagement. These formats shift engagement from discussion to co-creation and validation, enabling stakeholders to collectively build and test the foundations of interoperable AI governance. The goal is not only to exchange ideas, but to produce working governance solutions in real time.

Effective AI governance is emerging where policy, technical standards, and verification mechanisms converge. A first example is the risk-based regulatory approach (e.g., the EU AI Act), which provides a structured way to classify systems and align obligations with risk levels. Its strength lies in creating clear accountability layers, though its effectiveness will depend on implementation tools that enable verification beyond documentation. Second, international normative frameworks such as the OECD AI Principles and UNESCO Recommendation on AI Ethics have successfully established shared baselines for responsible AI, facilitating global alignment at the principles level. Their limitation-and opportunity-is the transition toward operationalization. Third, technical standardization efforts (ISO/IEC, IEEE) are advancing common definitions, metrics, and lifecycle processes. These are essential building blocks for interoperable governance, particularly when linked to certification and audit practices. Fourth, emerging practices in industry-such as algorithmic auditing, model cards, and system impact assessments-are improving transparency and accountability. However, they remain largely static and self-reported, highlighting the need for more robust assurance mechanisms. A promising direction is the development of continuous assurance systems, where AI models generate auditable, real-time evidence of compliance and performance. This includes advances in secure computation, logging infrastructures, and privacy-preserving verification. Finally, regulatory sandboxes and cross-sector pilot programs are proving effective in testing governance approaches under real conditions, enabling iterative learning between regulators and innovators. The most impactful approaches share a common trajectory: moving from principles and ex-ante compliance toward dynamic, verifiable, and interoperable governance systems, where trust is continuously tested rather than assumed.