RAND Europe

The Global Dialogue on AI Governance should move beyond high-level consensus building and instead anchor itself in applied problem-solving for urgent, real-world crises. At present, much of global AI governance remains concentrated in abstract principles or technologically advanced contexts, leaving a gap in how AI can be operationalised in settings of energy insecurity, climate stress, and humanitarian fragility. The Dialogue is uniquely positioned to address this imbalance. First, it should prioritise sector-specific application pathways, particularly in energy systems. The ongoing global energy crisis—driven by geopolitical instability, climate transitions, and infrastructure fragility—demands governance frameworks that guide how AI can be deployed in grid optimisation, demand forecasting, and resilience planning. This includes establishing shared standards for safety, reliability, and accountability in high-risk, critical infrastructure contexts, where failures carry systemic consequences. Second, the Dialogue should embed AI governance within humanitarian and crisis-response settings, where the stakes are immediate and often life-critical. Current governance approaches insufficiently address how AI systems function in low-resource, high-risk environments such as conflict zones, disaster response, and displacement contexts. The Dialogue should therefore promote operational guidance for AI deployment under conditions of uncertainty, including safeguards for data protection, bias mitigation, and meaningful human oversight when institutional capacity is constrained. Third, it should enable deeper, context-sensitive resolutions rather than broad alignment. This means shifting from general interoperability discussions to developing implementable governance toolkits—for example, risk assessment protocols, procurement standards, and monitoring indicators tailored to specific sectors and geographies. These should be co-developed with countries facing acute structural challenges, ensuring that governance reflects lived realities rather than imported models. Ultimately, the Dialogue should achieve a transition from principle-setting to practice-shaping: enabling AI governance that is not only inclusive in voice, but effective in addressing the interconnected crises defining the current global landscape.

• AI capacity-building
• Protection and promotion of human rights
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI

The selected priorities reflect a need to shift global AI governance from abstract alignment to applied, high-impact implementation in contexts where risks are immediate and capacity is uneven. Safe, secure and trustworthy AI is a core priority given the increasing integration of AI into critical systems, particularly in energy, public services, and security domains. My work focuses on how AI intersects with critical national infrastructure, where failures can cascade across sectors. This requires robust assurance approaches-such as testing, evaluation, verification and validation, auditability, and incident monitoring-tailored to high-risk, real-world deployments rather than hypothetical risks. AI capacity-building is essential to address structural asymmetries in the ability of states to govern and deploy AI effectively. Many countries face constraints not only in technical capability but also in regulatory, institutional, and evaluative capacity. There is a clear need for practical toolkits, skills frameworks, and governance support that enable countries-particularly in the Global South-to engage meaningfully in AI governance and deployment, including in sectors such as energy resilience and humanitarian response. Protection and promotion of human rights remains central, particularly as AI systems are increasingly used in sensitive domains such as migration, welfare allocation, and information ecosystems. My work engages closely with rights-based frameworks, emphasising lifecycle governance, proportionality, and accountability. There is an urgent need to operationalise these principles in practice, including through impact assessments, redress mechanisms, and safeguards for democratic integrity. Finally, the social, economic, ethical, cultural, linguistic and technical implications of AI are critical to ensure that governance reflects diverse societal contexts. AI systems are not neutral; they embed assumptions that can exacerbate inequality or exclusion. Addressing these implications requires grounding governance in local realities, particularly in crisis-affected and resource-constrained settings, and ensuring that benefits are equitably distributed. Together, these priorities support a transition towards context-sensitive, operational AI governance that responds to interconnected global challenges.

A cross-cutting issue not fully captured by the listed themes is the convergence of AI with other emerging technologies in shaping systemic risk and crisis dynamics. AI is increasingly being deployed alongside advances in areas such as biotechnology, cyber capabilities, and advanced sensing, creating compound risk environments that extend beyond the scope of single-technology governance. For example, AI-enabled biological design, automated cyber operations, or the use of AI in scientific discovery can accelerate both beneficial and harmful capabilities. Current thematic areas address AI in isolation, but do not sufficiently capture how these intersections amplify risks, compress response times, and challenge existing regulatory boundaries. This is particularly relevant in crisis contexts-such as public health emergencies, conflict settings, or infrastructure disruptions-where converging technologies can be rapidly repurposed or misused. Governance approaches therefore need to account for dual-use pathways, rapid capability diffusion, and the difficulty of attribution and control in such environments. In my work, I have engaged with AI governance at the intersection with other frontier technologies, which has highlighted the practical difficulty of applying existing regulatory and risk frameworks to these converging domains. This includes questions around how to assess risk where system boundaries are unclear, or where capabilities evolve faster than governance processes. Addressing this gap would require developing integrated risk assessment approaches, cross-domain expertise within governance institutions, and mechanisms for anticipatory monitoring of emerging capabilities. Bringing this lens into the Dialogue would support more forward-looking governance that is better aligned with the evolving nature of technological risk, particularly in high-stakes and crisis-prone contexts.

Governance gaps across the selected thematic areas are already shaping how AI is deployed and managed across Europe, ASEAN, and India, revealing both shared pressures and region-specific dynamics. In Europe, advances in safe, secure and trustworthy AI and human rights-based governance—notably through the EU AI Act and the work of the Council of Europe—have established a strong normative and regulatory foundation. However, a key challenge lies in operationalisation: translating high-level requirements into sector-specific assurance practices, particularly in critical and crisis-relevant domains such as energy and public services. While these frameworks are influential globally, they require contextualisation to remain applicable beyond highly resourced regulatory environments. Across ASEAN, the primary challenge is capacity asymmetry. Although there is increasing engagement with AI governance, many countries face constraints in technical expertise, institutional capability, and enforcement. This affects the implementation of AI capacity-building, safety, and human rights protections in practice. There is a clear opportunity to develop regionally grounded governance models, supported by shared infrastructure, skills development, and practical toolkits that reflect diverse socio-economic contexts. In India, rapid advances in AI deployment—particularly in digital public infrastructure—highlight both opportunity and risk. There is strong momentum in scaling AI applications and building capacity, but governance gaps remain in consistent assurance mechanisms, transparency, and rights protections at scale. This underscores the need for implementable safeguards that can operate effectively in large, heterogeneous systems. Frameworks developed by organisations such as UNESCO and the Council of Europe provide important global reference points, particularly in embedding rights-based and ethical approaches. However, a key challenge across regions is ensuring these approaches are inclusive of Global South perspectives, adaptable to varying institutional capacities, and responsive to crisis contexts. Across all three regions, the central issue remains the transition from principles to practice, and the opportunity lies in developing context-sensitive, interoperable governance approaches that can be meaningfully applied across different settings.

The AI Dialogue can play a critical role in advancing international cooperation by shifting from a forum for high-level alignment to a platform that empowers Member States, civil society, and a broader range of stakeholders to act on shared challenges. First, it can strengthen inclusive participation and voice, ensuring that countries with differing levels of capacity—particularly from the Global South—are not only represented but able to shape outcomes. This includes enabling civil society, technical communities, and local actors to contribute grounded perspectives on how AI is experienced in practice, especially in crisis-affected and resource-constrained settings. Second, the Dialogue can support the co-development of practical, evidence-based governance tools. Rather than reiterating principles, it can facilitate the creation of operational artefacts—such as risk assessment templates, assurance approaches, procurement standards, and monitoring indicators—that Member States can adapt and implement. Building a shared evidence base on what works, including lessons from deployments in sectors such as energy, public services, and humanitarian response, would significantly strengthen cooperation. Third, it can act as a bridge between norm-setting and implementation, helping translate existing frameworks into actionable guidance. This includes fostering peer learning, capacity-building partnerships, and cross-regional collaboration, enabling countries to learn from each other's experiences in applying governance approaches under different conditions. Finally, the Dialogue can promote collective anticipation of emerging risks, particularly as AI evolves alongside other technologies. By convening diverse actors, it can support early identification of shared challenges and coordinate responses before risks materialise at scale. In this way, the AI Dialogue can move beyond consensus-building to become a mechanism for collective action—grounded in evidence, inclusive in participation, and focused on real-world implementation.

The AI Dialogue should build on existing normative, regulatory, and applied governance initiatives, while addressing the current gap between principle-setting and implementation. At the global level, frameworks such as the UNESCO Recommendation on the Ethics of AI and emerging instruments from the Council of Europe provide a strong rights-based and lifecycle governance foundation. In Europe, the EU AI Act represents a significant advance in operationalising risk-based regulation. These should be complemented by more applied initiatives, including AI safety and evaluation efforts, sectoral regulatory sandboxes, and capacity-building programmes across ASEAN and India. The Dialogue should also connect with multi-stakeholder and technical communities, including standards bodies, open-source ecosystems, and research networks working on AI evaluation, assurance, and risk assessment. Existing partnerships in areas such as digital public infrastructure, humanitarian innovation, and energy system resilience offer practical entry points where AI governance is already being tested under real-world constraints. From my work across Europe, ASEAN, and India, a consistent gap is not the absence of frameworks, but the lack of contextualised, implementable guidance. The added value of the AI Dialogue would therefore be threefold. First, it can act as a translation layer, turning high-level principles into operational tools—particularly through structured Testing, Evaluation, Verification and Validation (TEVV) approaches. This includes pre-deployment testing against defined risk thresholds; evaluation of system performance, bias, and robustness across contexts; verification that systems meet specified technical and regulatory requirements; and ongoing validation in real-world conditions through monitoring, audits, and incident reporting. Embedding TEVV across the lifecycle would support continuous assurance, rather than one-off compliance checks. Second, it can support capacity-aligned cooperation, linking countries with differing capabilities through peer learning, joint pilots, and shared evidence on deployment in critical and crisis-relevant domains. Third, it can convene cross-domain expertise to address emerging risks at the intersection of AI and other technologies, ensuring governance remains forward-looking. In doing so, the Dialogue can become a mechanism that connects existing efforts while enabling coordinated, action-oriented implementation at scale.

The AI Dialogue should be structured to enable meaningful contribution from governments, industry, technical communities, and civil society, while ensuring outputs are action-oriented and implementable. First, establish multi-track engagement formats. Alongside formal intergovernmental discussions, a dedicated Track 2 should convene industry, technology experts, and civil society to address practical challenges such as AI deployment risks, governance in crisis contexts, and sector-specific implementation. This would allow more candid exchange and faster iteration of solutions, which can then inform formal negotiations. Second, embed capacity-building sessions as a core component, not a parallel activity. These should be hands-on and tailored—covering areas such as risk assessment, procurement practices, and human rights impact assessments. Pairing countries with different levels of capacity in peer-learning formats would help translate principles into practice. Third, adopt more inclusive and interactive dialogue formats, such as fishbowl discussions, which allow a rotating set of participants—particularly from underrepresented regions and civil society—to directly engage in core conversations. This can help surface grounded insights from humanitarian, local governance, and community perspectives that are often missing in formal settings. Fourth, structure the Dialogue around problem-focused working groups, for example on energy systems, public services, or humanitarian response, with clear deliverables such as toolkits, indicators, or pilot initiatives. This would anchor discussions in real-world use cases. One additional idea would be to introduce "implementation labs" within the Dialogue—small, time-bound working sessions where mixed stakeholder groups co-develop concrete governance artefacts (e.g. a sector-specific risk framework or a monitoring approach for AI in crisis response). These labs would ensure the Dialogue produces tangible outputs, not only shared understanding. Overall, the Dialogue should balance inclusivity with structure, enabling broad participation while driving practical, evidence-based outcomes.

The lived experience of AI is already widespread—across welfare systems, labour markets, information ecosystems, and public services—yet the voices of those most affected remain underrepresented in global governance discussions. This includes civil society organisations, frontline practitioners, and communities directly impacted by AI deployments, particularly in low-resource and crisis-affected settings. Their perspectives are critical for understanding how systems function in practice, where harms emerge, and what meaningful accountability looks like. A second gap concerns the Global South, which is often positioned as a recipient of AI systems or governance models developed elsewhere. There is a need to shift towards recognising countries and communities as co-creators of governance approaches, drawing on local innovation, policy experimentation, and contextual knowledge. This includes elevating regional institutions, local research networks, and public sector actors shaping digital transformation on the ground. There are also persistent age and gender gaps in AI governance spaces. Youth perspectives are often excluded despite being disproportionately affected by long-term technological change, while gender imbalances continue to shape whose priorities are reflected in system design and governance. Similarly, linguistic diversity remains limited, constraining participation from non-English-speaking communities. In my work across different regional contexts, engagement with public sector institutions and local stakeholders has highlighted how governance frameworks often overlook implementation realities and lived experience, particularly outside highly resourced environments. This reinforces the need for more grounded and inclusive approaches. To address these gaps, the AI Dialogue could adopt participatory mechanisms such as structured civil society forums, regional consultations, and community-led evidence inputs. Providing support for participation—through funding, translation, and capacity-building—would be essential. Embedding these perspectives systematically would help ensure that AI governance reflects not only global principles, but the realities of those most affected.

Meaningful engagement in the AI Dialogue will depend on formats that move beyond formal statements and enable interactive, inclusive, and context-rich participation. First, multilingual engagement should be embedded throughout. This goes beyond translation of final outputs to include real-time interpretation, multilingual breakout sessions, and the ability to submit inputs in diverse languages. This is essential to ensure that participation is not limited by language and that perspectives—particularly from the Global South—are expressed with nuance. Second, incorporating visual and arts-based participatory methods can open up new forms of engagement. Storytelling, design labs, and visual mapping exercises (e.g. illustrating how AI affects a community or sector) allow participants—especially those outside technical or policy communities—to communicate lived experiences and risks in accessible ways. This is particularly valuable in humanitarian and public service contexts, where impacts are complex and not easily captured through formal interventions. Third, scenario-based simulations and crisis exercises could be used to explore governance challenges in real time. Mixed stakeholder groups (governments, industry, civil society) could work through simulated incidents—such as an AI-related infrastructure failure or misinformation surge—helping to surface coordination gaps and practical governance needs. Fourth, rotating small-group formats such as fishbowl discussions or "open floor" exchanges can ensure broader participation while maintaining focus. These allow dynamic entry of participants, including underrepresented voices, into core discussions. Fifth, digital participation platforms—including asynchronous inputs, polling, and collaborative drafting tools—can extend engagement beyond those physically present and enable iterative contributions. Finally, introducing co-creation or "implementation labs" would allow participants to jointly develop concrete outputs, such as sector-specific guidance or monitoring approaches, within the Dialogue itself. Together, these formats can foster a Dialogue that is not only inclusive, but experiential, participatory, and oriented towards practical outcomes.

Drawing on my work across policy design, evaluation, and implementation, several initiatives illustrate how AI governance can move from principle to practice across different contexts. At the global level, the UNESCO Recommendation on the Ethics of AI provides a comprehensive, rights-based framework with practical tools such as readiness assessments and capacity-building support. Similarly, instruments developed by the Council of Europe-including its emerging AI convention and risk and impact assessment methodologies-offer operational approaches to embedding human rights across the AI lifecycle. From a development and implementation perspective, the World Bank has advanced digital public infrastructure and GovTech approaches, supporting countries to integrate AI into public services while strengthening institutional capacity and safeguards. These efforts demonstrate how governance can be embedded within broader digital transformation agendas. At the national level, the UK AI Safety Institute represents an important step towards applied AI assurance, including evaluation methodologies, testing approaches, and collaboration with international partners on frontier risks. In parallel, regulatory frameworks such as the EU AI Act illustrate how risk-based classification and compliance mechanisms can be formalised in law. Additional practices include regulatory sandboxes (e.g. in Singapore and the UK), which allow controlled experimentation, and multi-stakeholder standards development through organisations such as ISO and OECD, which help align technical and policy communities. In my experience, these initiatives are most effective when they are adapted to local contexts and linked to implementation mechanisms-including skills development, procurement practices, and monitoring systems. While I do not explicitly reference the Global Digital Compact here, as it underpins the intent of the Dialogue itself, these examples demonstrate the range of actionable approaches that the AI Dialogue can connect, scale, and contextualise across regions.