Institute of Tax AI Governance

A successful first Global Dialogue on AI Governance should deliver outcomes that are both principled and implementable—setting the tone for sustained global coordination rather than a one-off convening. First, success would mean achieving clear alignment on core governance principles—particularly around transparency, accountability, explainability, and auditability of AI systems. Not just abstract agreement, but language that can be adopted across jurisdictions and institutions. Second, the Dialogue should produce at least one practical output, such as a draft framework, model policy, or implementation roadmap. For example, an outline of minimum governance controls for high-risk AI systems, or a baseline for audit-ready AI deployment. This ensures the conversation translates into real-world application. Third, cross-sector engagement is critical. A successful outcome would involve meaningful participation and buy-in from regulators, industry leaders, academia, and international organizations—demonstrating that governance is not being developed in silos. Fourth, the Dialogue should establish a continuity mechanism—such as a working group, annual forum, or task force—to carry forward the work. Governance in this space requires iteration, not finality. Finally, success would be reflected in institutional commitment: where participating organizations signal willingness to pilot, adopt, or further develop the outputs of the Dialogue within their own systems. In essence, the measure of success is whether the Dialogue moves AI governance from conversation to coordination, and from coordination to credible implementation.

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

Our selections reflect a focus on moving AI governance from principle to practical implementation within institutional and cross-border contexts. Safe, secure and trustworthy AI is foundational. In regulated environments such as tax and compliance systems, trust must be engineered through robust governance controls, risk management frameworks, and assurance mechanisms that make AI systems reliable in practice-not just in theory. Transparency, accountability, and human oversight are central to achieving audit-ready AI. Institutions require systems that can be explained, reviewed, and defended-particularly where AI outputs inform regulatory decisions, enforcement actions, or taxpayer outcomes. This aligns with the need for clear documentation, traceability, and defined human-in-the-loop processes. Interoperability of governance approaches is critical given the cross-border nature of modern regulatory systems. Divergent frameworks increase friction and risk. There is a strong need for alignment across jurisdictions to enable consistent standards, facilitate cooperation among authorities, and support multinational adoption of AI governance practices. AI capacity-building ensures that governance frameworks can be effectively implemented. Many institutions face gaps in technical, legal, and operational readiness. Targeted training, certification, and institutional support are essential to translate governance principles into sustained practice. Collectively, these priorities emphasize credible implementation, institutional readiness, and global coordination-key conditions for advancing responsible and scalable AI governance.

Yes-several cross-cutting issues are not fully captured by the listed themes, particularly those that determine whether AI governance can be operationalized and sustained in practice. First, AI assurance and auditability remain underdeveloped. Beyond transparency, there is a need for standardized methods to test, validate, and continuously monitor AI systems. This includes independent assurance frameworks, audit trails, and mechanisms to evidence compliance in high-stakes environments. Second, model risk management and lifecycle governance are critical. AI systems evolve over time, and governance must extend across the full lifecycle-from design and training to deployment, monitoring, and decommissioning. Issues such as model drift, retraining, and version control require structured oversight. Third, regulatory readiness and enforcement capacity deserve greater focus. Many jurisdictions are advancing principles, but lack the institutional tools, technical expertise, and enforcement mechanisms to supervise AI effectively. Bridging this gap is essential for credibility. Fourth, data governance and provenance are increasingly central. Questions around data quality, lineage, consent, and cross-border data flows directly affect the reliability and legality of AI systems. Finally, governance of AI use in public sector decision-making is an emerging priority. Where AI informs taxation, benefits, or enforcement actions, there must be heightened safeguards to ensure fairness, due process, and contestability. These issues cut across sectors and jurisdictions, and addressing them is key to ensuring that AI governance frameworks are not only well-designed, but also enforceable, auditable, and trusted in real-world applications.

Governance gaps in the selected areas are already shaping outcomes across tax administration and broader compliance functions, particularly in cross-border and emerging-market contexts. A key challenge is the absence of standardized assurance and audit frameworks for AI systems. Institutions are adopting AI for risk assessment, enforcement, and taxpayer services, but often without clear mechanisms to validate outputs or demonstrate compliance. This creates exposure to errors, disputes, and diminished trust—especially where decisions affect financial obligations or legal rights. Another challenge is limited institutional capacity. Many authorities and organizations lack the technical and governance expertise required to implement effective oversight, including model validation, monitoring, and documentation. This is compounded by fragmented approaches across jurisdictions, where inconsistent standards hinder coordination and increase compliance burdens for multinational entities. Interoperability gaps further affect cross-border systems. Divergent governance frameworks make it difficult to align processes, share data responsibly, or rely on AI-driven outputs across jurisdictions. This is particularly relevant in international tax, where cooperation and consistency are essential. At the same time, there are significant opportunities. The current stage of adoption allows institutions to embed governance frameworks early—designing AI systems that are transparent, auditable, and aligned with regulatory expectations from the outset. There is also strong potential for capacity-building initiatives, including training and certification, to accelerate readiness across both public and private sectors. Finally, increasing global attention to trustworthy AI creates momentum for collaboration. Institutions that invest now in robust governance, assurance mechanisms, and cross-border alignment will be better positioned to lead in responsible AI deployment and to build durable public confidence.

The AI Dialogue can play a catalytic role by moving international cooperation from fragmented discussions to structured, sustained coordination. First, it can serve as a convening platform for alignment, bringing together regulators, industry, academia, and international organizations to develop a shared understanding of governance priorities. This is particularly important in reducing divergence across jurisdictions and establishing common reference points for policy and practice. Second, the Dialogue can support the development of globally relevant frameworks and baseline standards. Rather than duplicating efforts, it can synthesize existing initiatives into practical guidance—such as minimum governance controls, auditability benchmarks, and risk classification approaches that can be adapted across regions. Third, it can act as a bridge between policy and implementation. Many global discussions remain high-level; the Dialogue can focus on translating principles into operational tools, including model governance templates, assurance methodologies, and institutional readiness frameworks. Fourth, the Dialogue can enable capacity-building and knowledge exchange, particularly for jurisdictions and sectors with limited resources. Structured learning initiatives, peer exchanges, and technical support can help close gaps in expertise and enforcement capability. Fifth, it can establish continuity mechanisms—such as working groups or task forces—to ensure that cooperation is iterative and responsive to technological developments, rather than episodic. Ultimately, the value of the AI Dialogue lies in its ability to create a coordinated ecosystem where governance approaches are not only aligned in principle, but also interoperable in practice—supporting trust, scalability, and responsible deployment of AI across borders.

The AI Dialogue should build on existing global efforts that have already advanced principles, policy coordination, and technical standards. Key initiatives include the OECD AI Principles and related work on trustworthy AI; the UNESCO Recommendation on the Ethics of AI; the G7 Hiroshima AI Process; and the G20 discussions on digital economy and AI governance. In addition, the European Union AI Act provides an emerging regulatory model, while standards bodies such as ISO and NIST (through its AI Risk Management Framework) contribute technical guidance. Multi-stakeholder platforms like the Global Partnership on AI also play an important role in research and collaboration. The added value of the AI Dialogue lies in its ability to connect and operationalize these efforts. While many of these initiatives provide high-level principles or region-specific rules, there remains a gap in translating them into globally interoperable, implementation-ready frameworks. The Dialogue can serve as a coordination layer, aligning outputs across these bodies and reducing fragmentation. It can also focus on practical tools—such as audit methodologies, governance templates, and assurance mechanisms—that institutions can adopt regardless of jurisdiction. Importantly, the Dialogue can create space for sector-specific application, particularly in regulated areas like tax and compliance, where existing global initiatives are less granular. It can also strengthen capacity-building pathways, ensuring that emerging and developing jurisdictions are not left behind in implementation. In essence, the AI Dialogue should not duplicate existing work, but rather integrate, translate, and extend it—bridging the gap between global principles and credible, real-world deployment.

Different stakeholders can contribute most effectively when their roles are clearly defined and tied to implementation outcomes. Governments and regulators should provide policy direction, identify priority risk areas, and share regulatory experiences, including supervisory challenges and enforcement considerations. Industry and technology providers should contribute practical insights on system design, deployment, and risk management, including lessons from real-world implementation. Academia and research institutions can support with evidence-based analysis, evaluation methodologies, and forward-looking research on emerging risks. Professional bodies and civil society can contribute perspectives on accountability, ethics, and public trust, while also supporting awareness and capacity-building initiatives. International organizations can help align efforts across jurisdictions and ensure coherence with existing global frameworks. In terms of format and structure, the Dialogue should be designed for continuity and practical output: • A plenary track to set strategic direction and align on key themes. • Thematic working groups focused on priority areas (e.g., auditability, risk classification, cross-border interoperability), tasked with producing concrete outputs. • An implementation track, where institutions share case studies, pilot projects, and operational challenges. • A capacity-building stream, including training sessions, toolkits, and peer-learning exchanges. To ensure sustained impact, the Dialogue should operate as an ongoing platform, not a one-time event—supported by a light governance structure, clear deliverables, and periodic reporting. Ultimately, the effectiveness of the AI Dialogue will depend on its ability to combine inclusive participation with disciplined execution—ensuring that contributions from diverse stakeholders translate into actionable and globally relevant outcomes.

Several important voices remain underrepresented in global AI governance discussions, and their absence limits both legitimacy and practical effectiveness. First, Global South policymakers and regulators are often not sufficiently represented, despite being directly affected by imported AI systems and regulatory models. Their inclusion is essential to ensure that governance frameworks reflect diverse legal systems, institutional capacities, and socio-economic realities. Second, frontline public sector practitioners—such as tax administrators, compliance officers, and case handlers—are frequently overlooked. These are the individuals who interact with AI systems in operational settings and can provide critical insights into implementation challenges, system limitations, and real-world risks. Third, small and medium-sized enterprises (SMEs) and local technology providers are underrepresented compared to large multinational firms. Yet, they play a growing role in AI deployment and often face disproportionate compliance burdens due to limited resources. Fourth, independent assurance professionals—including auditors, risk specialists, and governance practitioners—are not consistently included, despite their central role in validating and monitoring AI systems. To address these gaps, the AI Dialogue should adopt a deliberate inclusion strategy: • Provide targeted participation pathways (e.g., sponsored seats, regional representation quotas, and fellowships) for underrepresented jurisdictions and practitioners. • Create practitioner-focused sessions where operational voices can contribute directly to discussions and outputs. • Support hybrid and accessible formats to reduce participation barriers, including virtual engagement and asynchronous contributions. • Establish structured feedback loops, ensuring that insights from these groups are integrated into final outputs and not treated as peripheral inputs. Broadening participation in this way will strengthen both the relevance and the credibility of global AI governance efforts.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional panels and adopt formats that prioritize interaction, co-creation, and real-world problem solving. First, scenario-based simulations can be highly effective. Participants are presented with realistic use cases—such as an AI-driven compliance system under audit or a cross-border data governance conflict—and asked to respond in real time. This encourages practical thinking and exposes gaps between policy and implementation. Second, co-creation labs (or governance design sprints) can bring together diverse stakeholders to develop specific outputs within a defined timeframe—such as a draft audit framework, risk classification model, or governance checklist. These sessions should be structured with clear deliverables and facilitation. Third, peer review clinics can allow institutions to present their existing AI governance approaches and receive structured feedback from experts across sectors. This promotes knowledge exchange while maintaining a focus on practical improvement. Fourth, implementation case rounds—short, focused presentations of real-world deployments followed by moderated discussion—can highlight what is working, what is not, and why. This helps ground the Dialogue in operational realities. Fifth, closed-door regulatory roundtables can enable candid discussions among policymakers and regulators on sensitive issues such as enforcement, supervision, and cross-border coordination. Finally, integrating digital collaboration tools—such as live polling, shared drafting platforms, and asynchronous input channels—can extend participation beyond the room and ensure broader engagement. These formats shift the Dialogue from passive listening to active participation, enabling stakeholders to collectively develop solutions that are practical, testable, and aligned with real-world governance needs.

Several existing policies, frameworks, and practices offer concrete pathways for effective AI governance, particularly where they move beyond principles into operational guidance. The National Institute of Standards and Technology provides a practical structure for identifying, assessing, and managing AI risks across the lifecycle. Its emphasis on governance, mapping, measurement, and management makes it adaptable across sectors. The European Union AI Act introduces a risk-based regulatory model, classifying AI systems by impact and imposing proportionate obligations. This approach offers a clear template for aligning regulatory oversight with system risk. The ISO and IEC standards (e.g., ISO/IEC 42001 on AI management systems) provide structured guidance for embedding governance within organizational processes, including documentation, accountability, and continuous improvement. From a policy perspective, the OECD AI Principles and the UNESCO Recommendation on AI Ethics establish widely recognized baselines for trustworthy AI, supporting international alignment. In practice, leading organizations are implementing model risk management frameworks adapted from financial services, incorporating validation, monitoring, and independent review functions for AI systems. Algorithmic impact assessments (AIAs) are also gaining traction as a tool to evaluate risks prior to deployment, particularly in public sector use. Additionally, sandbox environments-where regulators and innovators test AI systems under controlled conditions-are proving effective in balancing innovation with oversight. Collectively, these examples demonstrate that effective AI governance is best achieved through a combination of risk-based regulation, operational standards, and continuous assurance mechanisms that translate principles into enforceable and auditable practices.