YAIL Kajiado Hub

A successful first Global Dialogue on AI Governance would not be defined by consensus alone, but by clarity, alignment, and credible next steps. First, success would mean achieving a shared understanding of what "AI governance" actually requires in practice. Today, much of the debate remains abstract; translating high-level principles into operational expectations across jurisdictions would be a critical outcome. This includes clarity on roles (provider vs deployer), use-case level risk assessment, and what meaningful oversight looks like in real systems. Second, the dialogue should narrow key areas of fragmentation. Full harmonisation is unrealistic, but alignment on core concepts; such as high-risk classifications, transparency expectations, and accountability frameworks; would significantly reduce regulatory uncertainty and enable organisations to build governance that is both compliant and scalable. Third, success would involve recognising current gaps between policy ambition and organisational reality. Many institutions are not yet equipped to meet upcoming obligations. A credible outcome would therefore include practical pathways; guidance, reference frameworks, and examples; that help bridge this gap without stifling innovation. Finally, the dialogue should establish momentum beyond the event itself. This means creating mechanisms for ongoing collaboration between regulators, industry, and technical experts, as well as defining tangible follow-ups, whether through working groups, shared standards, or pilot initiatives. In short, success would not be measured by declarations, but by whether the dialogue moves AI governance from principles to practice, reduces uncertainty, and builds a foundation for sustained, global coordination.

Interoperability of governance approaches;

The current trajectory of AI regulation is increasingly fragmented, while AI systems themselves are inherently cross-border. Organisations are already navigating overlapping frameworks; such as the EU AI Act, UK principles-based guidance, and emerging regimes in the US and Asia; each with different definitions, risk classifications, and compliance expectations. Without a degree of interoperability, this creates duplication, legal uncertainty, and ultimately weakens governance, as efforts are spent reconciling frameworks rather than managing real risks. Interoperability does not require full harmonisation. Instead, it requires alignment on core building blocks; shared terminology, comparable risk categorisation, and mutually recognisable governance practices. For example, a use-case level risk assessment conducted under one framework should be interpretable and adaptable under another, rather than requiring complete rework. From an operational perspective, this is critical. Effective AI governance must be embedded into organisational processes; model development, deployment, vendor oversight, and monitoring. If governance requirements diverge too significantly across jurisdictions, organisations will either default to the lowest common denominator or struggle to scale responsible practices. Interoperability therefore becomes a prerequisite for both effective regulation and innovation. It allows organisations to build governance once, apply it globally, and focus resources on managing risks rather than navigating inconsistencies. In short, without interoperability, governance risks becoming fragmented and performative; with it, there is a realistic path toward scalable, accountable, and globally coherent AI oversight.

One cross-cutting issue that remains insufficiently captured is the gap between governance design and operational reality. Much of the current discourse focuses on defining principles, risk categories, and regulatory obligations. However, far less attention is given to how these are implemented within organisations; where governance must operate across complex systems, evolving use cases, and multiple lines of accountability. In practice, responsibilities are often fragmented, documentation is inconsistent, and key decisions; such as risk classification or acceptable use boundaries; are pushed to teams without the necessary expertise or oversight. A second emerging issue is the governance of general-purpose and agentic AI systems. These systems blur traditional distinctions between provider and deployer, and between tool and decision-maker. As capabilities evolve, governance models based on static classifications struggle to capture how systems are actually used, modified, and combined in real-world contexts. This raises questions around responsibility, traceability, and how to monitor systems whose behaviour may shift over time. A third gap is the increasing reliance on AI-to-AI interactions. As organisations deploy multiple systems that interact with each other, risks can emerge not from a single model, but from their combined behaviour. Existing governance approaches, which tend to assess systems in isolation, are not well suited to address these dynamics. Finally, there is a growing risk that governance becomes performative rather than effective. As regulatory pressure increases, organisations may prioritise documentation and formal compliance over genuine risk management. Without mechanisms to assess real-world outcomes, governance risks becoming a static exercise rather than a dynamic process. Addressing these issues requires moving beyond principles toward adaptive, operational governance that reflects how AI systems actually function in practice.

Governance gaps are increasingly visible across regions, reflecting a broader tension between rapidly advancing AI capabilities and slower, fragmented regulatory responses. In the EU, the AI Act provides a structured, risk-based framework, but uncertainty around implementation, guidance, and enforcement is already affecting organisations. Firms are preparing for high-risk obligations without full clarity, creating a risk of uneven or overly formalistic governance. In the UK, the more principles-based approach offers flexibility, but also places greater interpretative burden on firms. This can lead to inconsistent practices across the market, particularly where expectations are not yet clearly operationalised. In the US and other regions, the landscape remains more fragmented and sector-driven, with emerging state-level initiatives and voluntary frameworks. While this supports innovation, it also reinforces global inconsistency, particularly for multinational organisations deploying AI across jurisdictions. Across all regions, a common challenge is the gap between governance design and real-world implementation. As highlighted in my research, governance is often treated as a static compliance exercise, whereas AI systems evolve dynamically; through updates, integrations, and changing use cases. This creates blind spots in accountability, particularly as systems become more autonomous and interconnected. At the same time, this moment presents a significant opportunity. There is increasing convergence around core principles; risk-based approaches, transparency, and accountability; even if implementation differs. This creates a foundation for interoperability, allowing organisations to build governance frameworks that are adaptable rather than jurisdiction-specific. If leveraged effectively, these developments could shift AI governance from fragmented compliance toward a more coherent, globally aligned model that reflects how AI systems actually operate in practice.

The AI Dialogue can play a critical role as a bridge between regulatory ambition and practical implementation across jurisdictions. First, it can facilitate alignment on core governance building blocks. While full harmonisation is unlikely, the Dialogue can help establish shared reference points; such as common terminology, risk categorisation approaches, and baseline expectations for transparency and accountability. This would significantly improve interoperability and reduce fragmentation for organisations operating globally. Second, the Dialogue can act as a forum to translate principles into practice. By bringing together regulators, industry, and technical experts, it can surface how governance frameworks operate in real-world settings, identify gaps, and share workable models. This is particularly important given the current disconnect between policy design and organisational implementation. Third, it can support coordination on emerging challenges that transcend national boundaries; such as general-purpose AI, agentic systems, and AI-to-AI interactions. These areas require collective thinking, as unilateral approaches are unlikely to be effective. Fourth, the Dialogue can help build trust and reduce regulatory divergence over time. Regular engagement can enable jurisdictions to better understand each other's approaches, creating pathways for mutual recognition or at least compatibility of governance frameworks. Finally, it can ensure continuity. International cooperation on AI governance cannot be achieved through a single event; the Dialogue can establish ongoing mechanisms; working groups, shared guidance, or pilot initiatives; that sustain collaboration and adapt to technological developments. In essence, the AI Dialogue has the potential to move global AI governance from fragmented discussions toward coordinated, practical, and scalable approaches that reflect the cross-border nature of AI systems.

The AI Dialogue should build on existing international and multi-stakeholder initiatives that have already established foundations for cooperation, while addressing their current limitations. Key frameworks such as the OECD AI Principles, the G7 Hiroshima AI Process, and the Council of Europe's AI Convention provide important baseline alignment on risk-based governance, human oversight, and accountability. In parallel, regional approaches; including the EU AI Act and the UK's principles-based model; offer complementary perspectives on how these principles can be operationalised. Industry-led initiatives and sector-specific standards, particularly in financial services and technology, also contribute practical insights on implementation. However, these initiatives often operate in silos, with limited mechanisms for translating alignment at the principle level into interoperable governance in practice. The AI Dialogue can add value by connecting these efforts and focusing on operational convergence. Specifically, it can act as a coordination layer; mapping where frameworks align, where they diverge, and how organisations can navigate these differences without duplicating effort. It can also facilitate the development of shared reference models; for example, use-case level risk assessment approaches, documentation standards, or governance structures; that can be adapted across jurisdictions. In addition, the Dialogue can serve as a platform to capture lessons from real-world deployments and incidents, ensuring that governance evolves in response to how AI systems are actually used. This includes creating space for regulators and industry to jointly assess emerging risks and refine expectations accordingly. Ultimately, the added value of the AI Dialogue lies in moving from fragmented initiatives toward a more connected ecosystem; where principles, regulation, and practice reinforce each other rather than operate in parallel.

Different stakeholders should contribute based on their role in the AI ecosystem, with the Dialogue structured to ensure both diversity of perspectives and practical outcomes. Regulators should provide clarity on policy intent, emerging expectations, and areas of uncertainty, while remaining open to feedback on implementation challenges. Industry participants should contribute operational insights; how governance is applied in practice, where gaps exist, and what scalable approaches are working. Technical experts and researchers can help ground discussions in system capabilities and limitations, particularly as AI evolves. Civil society plays a critical role in highlighting societal impacts, ensuring that governance remains aligned with fundamental rights and public trust. To be effective, the AI Dialogue should move beyond a traditional conference format. A hybrid structure would be most valuable. First, plenary sessions can establish common framing; key risks, priorities, and areas of convergence. Second, smaller, thematic working groups should focus on specific issues such as interoperability, high-risk classification, or governance of general-purpose AI. These groups should be tasked with producing concrete outputs; for example, draft guidance, shared frameworks, or comparative mappings. Third, the Dialogue should incorporate case-based discussions. Real-world use cases and incidents provide a more effective basis for alignment than abstract principles, helping stakeholders converge on practical solutions. Finally, continuity is essential. The Dialogue should establish ongoing working mechanisms; such as standing groups or periodic reviews; to ensure that collaboration evolves alongside technological developments. In essence, stakeholders should not only exchange views, but co-develop practical approaches, with a structure designed to translate discussion into actionable outcomes.

Several important voices remain underrepresented in global AI governance discussions, particularly those closest to how AI systems are actually built, deployed, and experienced. First, operational practitioners within organisations; those responsible for implementing governance across legal, risk, compliance, and technology functions; are often missing. While policy discussions are well represented, there is less input from those translating requirements into practice. This contributes to the gap between regulatory design and real-world implementation. Second, perspectives from the Global South and smaller economies remain underrepresented. Much of the current agenda is shaped by a limited number of jurisdictions, yet AI systems are deployed globally. Including a broader range of regions is essential to ensure governance frameworks reflect diverse societal, economic, and cultural contexts. Third, end-users and affected communities are often included only indirectly. This is particularly important in sensitive domains such as employment, finance, and public services, where AI decisions have tangible impacts on individuals' lives. Their perspectives can highlight risks that may not be visible at the design or policy level. Fourth, there is limited representation of interdisciplinary expertise. AI governance is not solely a legal or technical challenge; it intersects with behavioural science, ethics, economics, and organisational design. Broader expertise can help anticipate unintended consequences and improve governance models. To address this, the AI Dialogue should adopt more inclusive participation mechanisms; targeted outreach to underrepresented regions, structured input from practitioners, and dedicated forums for affected communities. It should also create pathways for continuous engagement, rather than one-off participation. Strengthening these voices is not only a matter of representation, but a prerequisite for building governance that is both effective and globally relevant.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional panel discussions and adopt formats that reflect how AI governance is actually developed and tested in practice. First, case-based simulation exercises would be highly effective. Participants could work through realistic scenarios; for example, deploying a high-risk AI system across multiple jurisdictions; and collectively navigate governance challenges such as classification, accountability, and transparency. This allows stakeholders to move from abstract principles to operational decision-making. Second, cross-functional roundtables should be prioritised. Bringing together legal, risk, technical, and business stakeholders in small groups mirrors how governance decisions are made within organisations and helps surface practical tensions that are often overlooked in policy discussions. Third, structured "regulator–industry labs" could enable direct collaboration on emerging issues. These sessions would focus on co-developing guidance or testing governance approaches in real time, rather than simply exchanging views. Fourth, the Dialogue could incorporate sector-led contributions. For example, initiatives such as the EFAMA AI Task Force (European Fund and Asset Management Association), of which I am a member, bring together asset managers to develop practical tools and shared approaches to AI governance. Leveraging such groups can provide concrete, implementation-focused insights that complement broader policy discussions. Finally, iterative formats should be embedded. Rather than one-off exchanges, outputs from working sessions should be refined over time through follow-up sessions or digital collaboration platforms. Overall, the most effective formats will be those that prioritise interaction, real-world application, and co-creation; ensuring that the Dialogue produces not only discussion, but tangible progress toward operational and interoperable AI governance.

• Effective AI governance is not emerging from a single framework, but from a shift in mindset: from controlling technology to shaping how it interacts with society. One important approach is moving from static rules to adaptive governance. AI systems evolve, learn, and interact in ways that cannot always be anticipated. Governance must therefore be continuous, responsive, and capable of learning alongside the technology itself. Another is grounding governance in real-world impact rather than abstract categorisation. What matters is not what an AI system is, but what it does
• how it affects people, decisions, and opportunities. This shift helps ensure that governance remains focused on human outcomes, not technical labels. There is also growing recognition that governance must be shared. No single actor
• whether a government, company, or institution
• can fully oversee AI on its own. Effective approaches bring together diverse perspectives, combining global principles with local realities. Importantly, moments of crisis and public attention are becoming unexpected drivers of progress. When harmful or controversial uses of AI enter public consciousness, they often create rare alignment across stakeholders. These moments can be harnessed to accelerate cooperation and reinforce shared boundaries. Ultimately, the most promising approaches are those that treat governance not as a constraint, but as an enabler
• building trust, guiding innovation, and ensuring that AI develops in a way that reflects collective human values.