American University of Beirut - Global Health Institute

A successful first Global Dialogue on AI Governance would move beyond high-level discussion and deliver a clear, actionable foundation for collective progress. First, it should result in a shared global agenda and practical roadmap for AI governance, one that is flexible enough to be adapted across diverse national and regional contexts, yet concrete enough to guide policy development, implementation, and evaluation. This includes aligning on core priorities, timelines, and mechanisms for accountability. Second, success would include a clear identification and inclusion framework for stakeholders. This means not only recognizing traditional actors such as governments, academia, and the private sector, but also ensuring meaningful representation from civil society, affected communities, and underrepresented regions. Defining who needs to be at the table and how they engage would strengthen legitimacy and impact. Third, the dialogue should produce a mapped understanding of capacity-building needs across stakeholder groups. This includes technical literacy, regulatory expertise, ethical oversight, and institutional readiness. Identifying these gaps would allow for targeted investments and partnerships that enable effective participation in AI governance processes. Fourth, it should facilitate an open exchange on context-specific challenges, from regulatory fragmentation and resource constraints to political and economic barriers, paired with practical, collaborative solutions. This would help ensure that governance approaches are grounded in real-world constraints. Finally, success would be reflected in the systematic sharing of lessons learned and good practices from existing initiatives. Creating mechanisms to document and disseminate these insights would prevent duplication, accelerate learning, and support more coherent global progress. Ultimately, the dialogue should not end as a one-off event, but rather establish a sustained platform for collaboration, accountability, and continuous learning in AI governance.

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

Safe, secure and trustworthy AI is a priority because AI will only achieve meaningful impact if it is trusted by end-users, communities, institutions, and decision-makers. Trust is essential for the uptake of AI and its integration into existing workflows. For this reason, AI governance discussions should prioritize clear agreement on how safety, security, and trustworthiness can be achieved in practice. AI capacity-building is also critical. Policymakers, implementers, regulators, researchers, and other stakeholders need the skills and knowledge required to participate meaningfully in AI governance processes. This capacity-building should go beyond generic or one-size-fits-all approaches and instead be grounded in local contexts, needs, and constraints, ensuring that stakeholders can translate global principles into actionable, context-sensitive practices. Without this capacity, governance may remain limited to technical or high-level policy discussions that do not reflect real implementation needs. Greater consensus is needed on the most effective ways to strengthen AI-related capacities across different settings. Interoperability of governance approaches should also be prioritized. There is no shortage of AI governance resources, principles, and frameworks. However, many remain fragmented or difficult to align. Addressing gaps and conflicts between existing approaches would help avoid duplication, make better use of available resources, and support faster progress toward practical implementation. Ensuring that these frameworks are adaptable to different socio-political and cultural contexts will further enhance their relevance and usability across regions. Finally, transparency, accountability, and human oversight are essential because major gaps remain in how responsibilities are defined when AI systems cause harm or fail. Governance discussions should clarify how accountability is assigned, how AI systems are monitored, and how meaningful human oversight is maintained. This is especially important for ensuring that AI is

• One cross-cutting issue that should be more explicitly captured is the need for global coordination and multilateral cooperation on AI governance. AI governance cannot be addressed effectively by countries, institutions, or sectors working in isolation. AI systems often operate across borders, rely on shared digital infrastructure, and affect multiple sectors at the same time. This makes global coordination essential for developing governance approaches that are coherent, practical, and responsive to common risks. Multilateral cooperation is also important because countries are at different stages of AI development, adoption, and regulation. Some countries have advanced governance systems, while others are still building the required capacities, policies, and institutions. The AI Dialogue can help create a space where countries and stakeholders exchange experiences, identify shared challenges, and work toward common approaches without ignoring local needs and realities. This issue cuts across several listed themes, including safe, secure and trustworthy AI
• AI capacity-building
• interoperability of governance approaches
• and transparency, accountability, and human oversight. Without stronger coordination, AI governance may remain fragmented, with overlapping frameworks, inconsistent standards, and limited opportunities for shared learning. For this reason, global coordination and multilateral cooperation should be treated as a cross-cutting priority. They can help align governance efforts, reduce duplication, support countries with fewer resources, and promote more responsible AI development and use across different contexts.

In Lebanon and the wider MENA region, one of the most significant AI governance gaps is the limited capacity of stakeholders involved in governing AI. Policymakers, institutions, and other relevant actors often lack the technical and policy capacities needed to develop contextualized laws, regulations, policies, and frameworks that can guide the responsible uptake and implementation of AI. Strengthening these capacities would be an important first step toward more effective AI governance. A second major gap is the limited interoperability of governance approaches. Once the capacities of relevant stakeholders are strengthened, the next step should be to adapt existing global AI governance approaches and align them with local laws, regulations, policies, and implementation realities. This would be more efficient than developing new governance resources from the ground up. It would also help make better use of the many AI governance frameworks and principles that already exist. These gaps create clear challenges, but they also present important opportunities. The main challenge is that AI may be adopted without sufficient guidance, oversight, or accountability. This could limit trust and reduce the safe integration of AI into existing systems and workflows. The main opportunity is that Lebanon and the MENA region can build on existing global resources while ensuring that governance approaches are locally relevant. If stakeholder capacities are strengthened and governance approaches are adapted effectively, safe, secure and trustworthy AI becomes a more achievable outcome. In this process, transparency, accountability, and human oversight can be clearly defined and embedded within governance frameworks that reflect local needs and realities.

The AI Dialogue can play an important role in advancing international cooperation on AI governance by creating a space for knowledge exchange, consensus-building, and collaboration. First, it can support knowledge exchange between countries, institutions, researchers, policymakers, and other stakeholders working on AI governance. This is important because countries and sectors are at different stages of AI adoption and governance. Sharing experiences, challenges, and practical lessons can help stakeholders learn from one another and avoid repeating the same mistakes. Second, the AI Dialogue can facilitate convergence towards shared understanding of AI governance principles and their operationalization. While a growing number of ethical frameworks and regulatory initiatives exist, they often vary in scope, interpretation, and implementation. The Dialogue can help identify areas of alignment, clarify divergent perspectives, and support the translation of high-level principles into practical, interoperable governance approaches that respect national contexts while promoting global coherence. Third, it can enhance international cooperation by identifying opportunities for joint action, partnerships, and coordinated responses to common challenges and risks associated with AI. Given the transboundary nature of AI technologies and their societal implications, strengthened collaboration is essential to ensure that governance frameworks are effective, inclusive, and responsive. The Dialogue can therefore serve as a catalyst for collective initiatives in areas such as safety, transparency, accountability, capacity-building, and equitable access to AI benefits.

The AI Dialogue should build on existing initiatives that already support responsible AI in development and health. One important initiative is the Artificial Intelligence for Development (AI4D) Initiative, supported by the International Development Research Centre (IDRC) and the United Kingdom Foreign, Commonwealth and Development Office (UK FCDO). AI4D works with local AI innovation and policy communities to promote AI that is safe, inclusive, rights-based, ethical, and sustainable, particularly for addressing development challenges. Within AI4D, the Artificial Intelligence for Global Health Research Hub (AI4GH) is especially relevant. AI4GH supports the responsible design and use of AI to improve equitable health outcomes. It also connects people, ideas, and capacity-strengthening opportunities across different contexts. The AI Dialogue should also connect with the Global Health and Artificial Intelligence Network in the Middle East and North Africa (GHAIN MENA), a project under AI4GH. GHAIN MENA serves as the only regional hub in MENA focused on advancing responsible AI for global health, with a specific focus on sexual, reproductive, and maternal health. It supports six AI-driven health projects across the region through funding, mentoring, and capacity-building. It is also developing a contextualized checklist to guide responsible AI adoption for global health in low-resource settings, alongside a certificate to strengthen the capacity of health system stakeholders in MENA. GHAIN MENA operates alongside 3 other sister hubs across the globe to promote the responsible adoption of AI for global health, through different funding, mentoring, and capacity building activities. These initiatives provide practical experience in responsible AI implementation, governance, partnerships, and capacity-building. The added value of the AI Dialogue would be to connect these efforts more systematically, support knowledge exchange between them, and identify shared priorities for action. It could also help ensure that lessons from development and health initiatives inform wider international discussions on AI governance.

The AI Dialogue should be designed as a structured and participatory process that allows different stakeholders to contribute meaningfully to AI governance discussions. Rather than relying only on high-level panels, the Dialogue should create spaces where policymakers, institutions, implementers, practitioners, researchers, and civil society actors can share practical experiences, identify governance gaps, and discuss what is needed to make AI governance more applicable in real-world settings. A useful structure would be to organize the Dialogue in stages. First, separate stakeholder sessions could be held to allow each group to discuss the challenges they face from their own perspective. Policymakers could reflect on the development of laws, regulations, policies, and frameworks. Institutions could discuss readiness, internal policies, expertise, and compliance challenges. Implementers and practitioners could share what happens when governance guidance is applied in practice. This would help ensure that each group contributes concrete and experience-based input. Second, the Dialogue should include joint sessions that bring these groups together to compare perspectives and identify shared priorities. These sessions should focus on practical questions, such as what support is needed, where governance guidance is unclear, and what mechanisms could improve coordination across stakeholders. Third, the Dialogue should include case examples from different countries, sectors, and implementation settings. This would make the discussion more grounded and would allow participants to learn from both successes and challenges. Overall, the AI Dialogue should be action-oriented. Its structure should help participants move from general discussion toward practical recommendations, areas for cooperation, and concrete next steps for strengthening AI governance.

Several communities remain underrepresented in global discussions on AI governance, even though AI interventions are increasingly being co-designed with the people and communities they aim to serve. In particular, vulnerable and disadvantaged communities are still not sufficiently represented. These include rural communities, refugees, pregnant women, children, elderly populations, inmates, and other groups who may be affected by AI systems but have limited influence over how these systems are designed, governed, or implemented. This is a major gap because AI governance decisions can shape access to services, quality of care, safety, accountability, and trust. When these communities are not included, governance approaches may fail to reflect their needs, risks, and lived realities. This can make AI systems less relevant, less trusted, or less safe for the people most affected by them. These communities could be included through more deliberate and structured engagement. This may include consultations, community advisory groups, participatory workshops, and co-design processes that involve affected communities from the early stages of AI development and governance discussions. Their participation should not be symbolic. It should meaningfully inform priorities, safeguards, oversight mechanisms, and implementation decisions. Youth are also underrepresented. As frequent users of digital technologies, young people will live with the long-term consequences of today's AI governance decisions. Their perspectives can show how AI affects education, employment, health, privacy, safety, and social participation. Including youth can bring practical insights on everyday AI use and help shape governance approaches for future generations. Global AI governance discussions should therefore create dedicated spaces for underrepresented communities and the organizations that work closely with them. This would help ensure that AI governance is not only shaped by technical, institutional, or policy perspectives, but also by the people and communities most likely to experience its benefits and risks.

The AI Dialogue should use engagement formats that allow participants to share practical experiences, learn from each other, and work together on concrete governance challenges. One useful format would be stakeholder-specific roundtables. Policymakers, institutions, implementers, practitioners, researchers, private sector actors, and community representatives could first discuss the challenges they face from their own perspective. These discussions could then inform broader joint sessions. Case-based discussions would also be valuable. Participants could examine real or realistic examples of AI implementation and discuss the governance questions that arise, such as safety, accountability, transparency, and human oversight. This would help move the Dialogue beyond general principles and toward practical application. The Dialogue should also include community listening sessions. These sessions would create space for underrepresented communities to share their concerns, expectations, and experiences related to AI. This is important because governance discussions should include the people and communities most affected by AI systems. Multi-stakeholder problem-solving workshops could help participants work together on specific governance gaps, such as capacity-building, interoperability of governance approaches, and accountability. These workshops should focus on identifying practical steps and areas for cooperation. Finally, cross-regional exchange sessions could allow countries and regions at different stages of AI governance to share lessons, challenges, and useful approaches. This would support knowledge exchange and help avoid duplication.

From a global health perspective, effective AI governance should build on practical policies, platforms, and approaches that make AI safe, accountable, and useful for health systems. One example is the use of Software as a Medical Device (SaMD) regulations as a foundation for governing AI-enabled health tools. These regulations can help ensure that AI systems used in clinical decision-making meet safety and quality requirements before and after deployment. However, they need to be adapted to address AI systems that change over time. A second example is post-market surveillance for AI-enabled health technologies. AI systems should not only be assessed before approval. They should also be monitored after deployment to identify safety issues, performance drift, bias, or unintended consequences. Predetermined Change Control Plans can support this by defining in advance which AI system changes are allowed and how they will be controlled. Another useful approach is the creation of formal coordination mechanisms, such as AI councils or cross-ministerial working groups. These can help reduce fragmentation between health ministries, regulators, data protection authorities, and other institutions. The AI Dialogue should also build on regulatory sandboxes, which allow innovative AI health technologies to be tested under controlled conditions. This can generate real-world evidence while protecting patients and communities. Finally, global health governance can benefit from platforms such as global regulatory networks, communities of practice, early-warning systems for AI adverse events, and public directories of registered AI health solutions. These mechanisms can support knowledge exchange, transparency, accountability, and coordinated action across countries. The attached HealthAI report emphasizes that the main barriers to responsible AI in health are governance-related, not only technological, and that stronger coordination, capacity-building, and lifecycle oversight are needed.