Levuka Venture Lab

The success of the first Global Dialogue on AI Governance will depend on whether it moves beyond principles and begins to address how governance functions at the level of decision-making systems. Most current discussions focus on model behavior, ethics, and regulatory alignment. These are important, but they do not fully address a structural question that is emerging across institutions: whether human authority remains clearly aligned with decisions increasingly influenced by artificial intelligence. As AI systems become embedded across operational, financial, and compliance workflows, they begin to influence outcomes that carry real economic and institutional consequences. These systems often operate across business units, organizations, and jurisdictions, while accountability structures remain fixed within traditional governance frameworks. This creates a growing gap between where decisions are influenced and where responsibility resides. A successful dialogue would therefore introduce greater clarity around governance architecture. Specifically, it would: - Distinguish between technical system governance and institutional authority structures - Encourage mapping of decision ownership for AI-influenced outcomes - Promote the development of clear escalation and override mechanisms - Address cross-jurisdictional and platform-level dependencies that affect institutional control The objective should not be to prescribe uniform solutions, but to ensure that institutions can clearly identify where governance structures may no longer align with the systems they oversee. Establishing a shared understanding of this structural challenge will be critical to enabling effective and accountable AI governance globally.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• AI capacity-building

AI systems are no longer confined to isolated use cases. They are increasingly embedded across operational, financial, and compliance workflows, where they influence decisions that carry real economic and institutional consequences. This shift is what informs the selected priorities. The broad category of social, economic, and technical implications remains important, but these implications are now shaped by how AI systems function inside real decision environments, not just how they are designed or evaluated. Interoperability of governance approaches is critical because these systems often operate across organizational and jurisdictional boundaries. Decision influence does not remain contained within a single framework, which creates complexity in maintaining consistent oversight and responsibility. AI capacity-building is also essential, particularly at the institutional level. Many organizations are equipped to assess model performance, but are less prepared to evaluate how governance structures must adapt as AI becomes embedded in decision systems. This includes understanding how authority is assigned, how escalation occurs, and how accountability is maintained. Taken together, these priorities reflect a need to ensure that governance frameworks remain aligned with how AI systems are actually deployed and used in practice.

One cross-cutting issue that is not fully captured is the alignment between institutional authority and AI-influenced decision systems. As artificial intelligence becomes embedded across workflows, it increasingly influences decisions that carry financial, operational, and regulatory consequences. These systems often operate across multiple business units, platforms, and jurisdictions simultaneously. However, institutional accountability structures remain largely fixed within traditional governance models. This creates a structural condition in which decision influence becomes distributed, while responsibility remains centralized and formally assigned. The result is not necessarily a failure of technology, but a misalignment within governance architecture. In practice, this can lead to ambiguity around decision ownership, unclear escalation pathways, and difficulty in maintaining effective oversight when automated systems operate at scale. This issue cuts across all thematic areas, including risk management, accountability, interoperability, and capacity-building. Without addressing how authority is allocated and exercised over AI-influenced decisions, efforts in these areas may remain incomplete. An important emerging priority is therefore the development of frameworks that help institutions assess whether their governance structures remain aligned with the systems they oversee.

AI deployment is accelerating across sectors, particularly in financial services, telecommunications, and public sector administration, where systems are increasingly embedded in operational and compliance workflows. In these environments, governance gaps are becoming more visible. AI systems are influencing decisions related to risk assessment, customer screening, pricing, and operational allocation, often across multiple business units and, in some cases, across jurisdictions. However, authority and accountability structures have not always evolved at the same pace. The most significant challenge is maintaining clear decision ownership. As AI systems operate across workflows and platforms, it becomes less clear which institutional actors hold responsibility for outcomes, particularly when decisions are influenced by systems that are integrated across organizational boundaries. This creates difficulty in defining escalation pathways, applying consistent oversight, and ensuring that decisions remain aligned with institutional risk tolerance and regulatory expectations. In cross-border environments, these challenges are amplified by differences in regulatory frameworks and reliance on external platforms. At the same time, this presents a clear opportunity. Institutions that proactively examine how authority is allocated over AI-influenced decisions will be better positioned to deploy these systems responsibly while preserving accountability and operational control. Strengthening governance architecture, including decision ownership mapping and escalation structures, can enable more confident adoption of AI across critical workflows. As deployment continues to scale, the ability to align governance structures with AI-influenced decision environments will become a key differentiator for institutional resilience and trust.

The AI Dialogue can play an important role in advancing international cooperation by helping to establish a shared understanding of how AI systems operate across institutional and jurisdictional boundaries. As artificial intelligence becomes embedded in operational and decision-making systems, its influence increasingly extends across organizations, sectors, and countries. However, governance frameworks remain largely developed within national or institutional contexts. This creates challenges in maintaining consistent oversight, accountability, and coordination when AI-influenced decisions cross these boundaries. The Dialogue can contribute by shifting part of the focus from principles and high-level alignment toward how governance functions in practice. In particular, it can help highlight the need for greater clarity around decision ownership, escalation mechanisms, and institutional responsibility in cross-border environments. It can also support the development of a shared language for describing how authority and accountability operate in AI-influenced systems. This would improve interoperability between governance approaches, not only at the regulatory level, but at the level of institutional decision-making. By facilitating structured exchanges between governments, institutions, and technical actors, the Dialogue can help surface common patterns and challenges that are emerging across jurisdictions. Strengthening international cooperation will depend not only on aligning standards, but on ensuring that governance structures remain coherent as AI systems operate across increasingly interconnected environments.

A number of existing initiatives provide important foundations for international cooperation on AI governance, including efforts focused on responsible AI principles, technical standards, and regulatory coordination. These initiatives have contributed to establishing shared norms around safety, transparency, and accountability. The AI Dialogue can build on this work by connecting these efforts more directly to how AI systems are deployed within institutional decision environments. While many existing initiatives focus on model behavior, data governance, and ethical considerations, there is an opportunity to strengthen focus on governance architecture. This includes how authority is allocated, how decision ownership is defined, and how escalation and oversight function as AI systems operate across organizations and jurisdictions. The added value of the Dialogue would be in bridging this gap. Specifically, it can: Facilitate exchange of practical experiences from institutions deploying AI across sectors Surface recurring governance challenges that are not fully addressed by existing frameworks Encourage alignment not only of principles, but of how governance structures operate in practice By connecting high-level frameworks with operational realities, the Dialogue can help ensure that existing initiatives remain relevant as AI deployment continues to scale. This would strengthen the ability of institutions to maintain accountability and effective oversight in increasingly complex, cross-border AI environments.

Different stakeholders can contribute to the AI Dialogue most effectively when their roles reflect how artificial intelligence is actually developed, deployed, and governed across institutions. Governments and regulators can contribute by providing policy direction and identifying areas where coordination across jurisdictions is required. Private sector institutions can offer insights from real-world deployment of AI systems within operational, financial, and compliance workflows, where governance challenges often become visible in practice. Academic and technical communities can support analytical rigor and help develop shared frameworks, while civil society can ensure that broader societal implications remain part of the discussion. To support meaningful contribution, the Dialogue should be structured to connect these perspectives rather than treat them in isolation. This can be achieved by: -Framing sessions around shared governance challenges rather than stakeholder categories -Encouraging cross-sector participation within each session -Incorporating case-based discussions where multiple stakeholders examine the same scenario from different perspectives A structured approach that integrates policy, technical, and operational viewpoints will allow stakeholders to contribute from their areas of expertise while building a more coherent understanding of how AI governance functions in practice.

Global discussions on AI governance often underrepresent institutions that are actively deploying AI in operational environments but are not directly involved in shaping policy frameworks. This includes mid-sized financial institutions, regional telecom operators, public sector agencies, and organizations operating in emerging markets. These actors are frequently managing real-world implementation challenges, including cross-jurisdictional deployment, reliance on external platforms, and evolving regulatory expectations. Their perspectives are important because governance challenges often become visible first at the level of deployment, rather than at the level of policy design. To better include these voices, the Dialogue could: -Expand participation beyond large multinational technology firms and major economies -Incorporate regional and sector-specific sessions that reflect diverse operational environments -Create structured opportunities for practitioners to share deployment experiences and governance challenges Inclusion should focus not only on representation, but on capturing practical insights from those directly managing AI-influenced systems.

To foster meaningful and dynamic engagement, the Dialogue should incorporate formats that move beyond traditional panels and enable participants to engage with how governance operates in practice. One effective approach would be scenario-based workshops, where participants examine specific AI deployment cases and discuss how governance structures function in real environments. This allows stakeholders to engage with concrete challenges such as decision ownership, escalation pathways, and cross-border coordination. Structured roundtables that bring together policymakers, institutional operators, and technical experts can also help surface common patterns across sectors and jurisdictions, rather than maintaining siloed perspectives. The Dialogue could further benefit from iterative working sessions that continue across multiple stages, allowing participants to refine insights over time rather than limiting engagement to one-off discussions. Finally, mechanisms to capture and synthesize insights in a structured way will be important to ensure that contributions translate into actionable understanding.

A number of existing approaches provide important foundations for effective AI governance, particularly in areas such as risk management, transparency, and accountability. Regulatory frameworks and guidance have increasingly focused on model evaluation, data governance, and operational controls. In practice, many institutions have adopted internal practices such as model risk management frameworks, independent validation processes, audit mechanisms, and oversight committees to manage AI deployment. These approaches help ensure that systems meet defined performance and compliance standards. In addition, emerging practices around documentation, explainability, and human oversight have strengthened visibility into how AI systems function and how decisions are produced. These developments represent meaningful progress. However, as AI systems become embedded within operational and decision-making environments, an additional dimension is becoming increasingly important. Specifically, there is a need to ensure that governance approaches extend beyond how systems are evaluated to how decisions influenced by those systems are owned, escalated, and governed within institutions. In many cases, governance mechanisms exist at the system level, while authority structures governing decision outcomes remain less explicitly defined, particularly when AI systems operate across business units, platforms, or jurisdictions. Approaches that integrate technical governance with clear authority mapping, defined escalation pathways, and institutional accountability structures will be critical in addressing this gap. Strengthening this alignment will help ensure that existing governance practices remain effective as AI deployment continues to scale across complex institutional environments.