Decision Infrastructure Advisory

A successful first Global Dialogue on AI Governance would help establish a shared understanding that effective AI governance depends not only on regulatory principles, but also on the institutional capability required to implement accountability consistently in practice. Many jurisdictions already reflect accountability obligations through existing data protection law, administrative law, governance frameworks, and sector-specific regulation. In South Africa, for example, obligations relating to oversight, procedural fairness, governance accountability, and automated decision-making are already reflected through frameworks such as POPIA Section 71, PAJA, King IV, and the oversight role of the Information Regulator South Africa. However, operational capability for verification, auditability, enforcement coordination, and institutional oversight may still remain uneven across different contexts. An important outcome of the Dialogue would therefore be greater focus on practical governance implementation beyond high-level principles. This could include supporting the development of: * Verification and auditability standards * Institutional enforcement capability * Decision traceability mechanisms * Governance approaches that remain adaptable across different institutional and regulatory environments. Another important outcome would be stronger multilateral coordination without encouraging one-size-fits-all governance models. Jurisdictions are approaching AI governance through different pathways, including layered legal frameworks, strategy-led approaches, and coordination-based models. The Dialogue could therefore support interoperability and harmonisation while still allowing contextual adaptation. Success would also require meaningful inclusion of Global South perspectives, not only as implementation environments, but as contributors to governance thinking. Relational accountability perspectives reflected in concepts such as Ubuntu and Ujamaa may also contribute useful approaches emphasising collective responsibility, social legitimacy, and community impact. Ultimately, the success of the Dialogue would be reflected in whether it helps strengthen governance systems that are operationally sustainable, contextually adaptable, and capable of implementing accountability in practice.

• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Interoperability of governance approaches
• AI capacity-building

The selected priorities reflect a shared governance challenge emerging across multiple jurisdictions: the gap between formal AI governance frameworks and the operational capability required to implement accountability consistently in practice. Transparency, accountability, and human oversight are central because institutions are increasingly expected to explain, verify, and defend AI-influenced decisions, particularly in areas involving personal data, administrative decision-making, and access to services. Existing legal and governance frameworks in many contexts already create accountability obligations, although operational systems for verification, oversight, and enforcement remain uneven. In South Africa, for example, existing governance obligations are already reflected through frameworks such as POPIA Section 71, PAJA, King IV, and the oversight role of the Information Regulator South Africa. Together, these mechanisms reinforce accountability, procedural fairness, explainability, and oversight in decision-making environments involving personal data and automated processes. Interoperability of governance approaches is important because AI governance is increasingly shaped across jurisdictions rather than within isolated national systems. Different governance pathways are emerging globally, including layered legal frameworks, strategy-led approaches, and coordination-based models. Effective multilateral governance therefore requires harmonisation and interoperability while still allowing contextual adaptation. AI capacity-building remains critical, particularly where discussions focus predominantly on technical skills and infrastructure. Capacity must also include governance literacy, enforcement capability, auditability, traceability, and the ability to sustain accountability mechanisms in practice. Protection and promotion of human rights remains foundational because AI governance directly affects fairness, dignity, procedural justice, and the ability of individuals to understand or challenge decisions that affect them. Across these priorities, there is also an opportunity for Global South perspectives, including experiences emerging from Southern African contexts such as South Africa and Zimbabwe, to contribute meaningfully to governance thinking. Relational accountability perspectives reflected in concepts such as Ubuntu and Ujamaa may also contribute useful governance approaches emphasising collective responsibility, social legitimacy, and community impact.

One important cross-cutting issue that is not yet sufficiently reflected in many AI governance discussions is institutional operational readiness - specifically, whether institutions have the practical capability to implement, verify, and sustain accountability in AI-influenced decision-making environments. Many governance frameworks appropriately emphasise principles such as fairness, transparency, safety, and human rights. However, less attention is often given to the operational systems required to enforce those principles consistently in practice. This includes capability for: * Verification and auditability, * Decision traceability, * Evidentiary standards, * Enforcement coordination, and * Governance literacy within institutions. Without these systems, accountability may exist in principle but remain difficult to implement, enforce, or defend in practice, particularly in ways that sustain public trust and institutional legitimacy. Another emerging issue is the growing gap between the speed of AI adoption and the slower pace of institutional adaptation. In many jurisdictions, accountability obligations already exist through data protection law, administrative law, governance frameworks, and sector-specific regulation, even before AI-specific legislation is introduced. In South Africa, existing governance obligations relating to automated decision-making, procedural fairness, explainability, and oversight are already reflected through frameworks such as POPIA Section 71, PAJA, King IV, and the oversight role of the Information Regulator South Africa. These developments suggest that many AI governance challenges are already emerging through existing institutional and legal systems. There is also a need to address governance asymmetries between jurisdictions. While AI governance is increasingly shaped multilaterally, institutional capacity and influence over governance standards remain unevenly distributed. This raises important questions about equitable participation in shaping global governance frameworks. Finally, governance discussions could benefit from greater engagement with relational accountability perspectives emerging from Global South contexts, including concepts such as Ubuntu and Ujamaa, which emphasise collective responsibility, social legitimacy, and community impact in governance design.

Across many sectors in Southern African contexts, including South Africa and Zimbabwe, AI governance developments are revealing a growing gap between increasing accountability expectations and the institutional capability required to implement them consistently in practice. One of the most significant challenges is that AI systems are being integrated into decision-making environments faster than governance systems are adapting. In areas such as recruitment, credit scoring, education, fraud detection, public administration, and access to services, institutions are increasingly expected to explain, verify, and defend AI-influenced decisions. However, operational capability for auditability, traceability, enforcement, and oversight often remains uneven. This creates risks not only for regulatory compliance, but also for procedural fairness, public trust, and the ability of individuals to understand or challenge decisions that affect them. In South Africa, existing frameworks such as POPIA, PAJA, and King IV, together with the oversight role of the Information Regulator South Africa, already establish accountability obligations relating to fairness, explainability, procedural justice, governance oversight, and automated decision-making. POPIA Section 71 is particularly significant because it creates obligations relating to automated decision-making, rights protection, and meaningful oversight where personal data is processed through automated systems. In Zimbabwe, existing accountability obligations are also reflected through data protection legislation, alongside the development of a National AI Strategy. These developments suggest that governance discussions in both contexts are increasingly focused not only on regulatory design, but also on implementation capability, governance literacy, enforcement coordination, and institutional readiness. Another significant challenge is the risk of governance fragmentation as jurisdictions adopt different regulatory and policy approaches while operating within increasingly interconnected digital systems. This creates pressure for interoperability and multilateral coordination without relying on one-size-fits-all governance models. At the same time, important opportunities are emerging. AI governance discussions are creating momentum for strengthening institutional accountability systems, improving governance coordination, and enabling African perspectives to contribute more meaningfully to global governance thinking. Perspectives grounded in relational accountability, including Ubuntu and Ujamaa, may also strengthen governance approaches by emphasising collective responsibility, social legitimacy, and community impact.

The AI Dialogue can play an important role in advancing international cooperation by helping establish a shared understanding that effective AI governance depends not only on regulatory principles, but also on the institutional capability required to implement accountability in practice. One of its most valuable functions would be to reduce fragmentation in global governance approaches by enabling jurisdictions to exchange implementation experiences, operational challenges, and governance models across different legal, economic, and institutional contexts. As AI systems increasingly operate across interconnected digital environments, international cooperation will depend on interoperability, coordination, and mutual learning rather than regulatory uniformity. The Dialogue could also help bridge the gap between high-level governance principles and operational implementation. Many jurisdictions already have accountability obligations embedded within existing legal and governance frameworks, including data protection law, administrative law, corporate governance frameworks, and sector-specific regulation. In South Africa, this is reflected through mechanisms such as POPIA Section 71, PAJA, King IV, and the oversight role of the Information Regulator South Africa, which collectively reinforce accountability, procedural fairness, explainability, and governance oversight in decision-making environments. However, institutional capability for enforcement, verification, auditability, and oversight often remains uneven. International cooperation should therefore include practical collaboration on governance infrastructure, audit standards, institutional readiness, and enforcement coordination. Another important role for the Dialogue is ensuring that global governance discussions remain inclusive and multilaterally representative. Governance capacity and influence over global standards remain unevenly distributed across regions. The Dialogue therefore provides an opportunity to strengthen more equitable participation in shaping governance approaches, including contributions from Global South governance experiences and relational accountability perspectives such as Ubuntu and Ujamaa. Ultimately, the Dialogue can help strengthen trust, legitimacy, and policy coherence in global AI governance by supporting governance approaches that are operationally sustainable, contextually adaptable, and internationally interoperable.

Several existing policies, frameworks, and governance initiatives provide useful foundations for effective AI governance and could inform the work of the AI Dialogue. At the normative level, the UNESCO Recommendation on the Ethics of Artificial Intelligence and the OECD AI Principles establish important foundations for human rights, accountability, transparency, interoperability, and trustworthy AI governance across jurisdictions. These principles increasingly intersect with existing domestic governance obligations, including data protection, procedural fairness, explainability requirements, and institutional oversight mechanisms. At the operational level, the EU AI Act, the NIST AI Risk Management Framework, and ISO/IEC 42001 offer practical approaches to risk classification, governance processes, auditability, institutional oversight, and AI management systems. These frameworks are particularly valuable because they move beyond high-level principles toward implementation and organisational accountability. Regionally, the African Union Continental AI Strategy is significant because it recognises varying levels of digital readiness, institutional capability, and developmental context across African states. This is important because effective governance cannot assume uniform technological advancement, infrastructure, or governance capacity across all societies. Emerging coordination mechanisms such as AI councils, regulatory cooperation platforms, and recent G7 discussions on trustworthy and human-centered AI also demonstrate growing recognition that AI governance requires institutional coordination and multilateral cooperation, not only standalone regulation. However, these approaches should not be treated as universally transferable without contextual adaptation. Governance models developed within highly resourced institutional environments may not always translate seamlessly across different legal, social, economic, and infrastructural realities. The added value of the AI Dialogue would therefore be its ability to connect these initiatives while supporting interoperability, institutional readiness, operational accountability, and more equitable participation in shaping global AI governance approaches.

Different stakeholders can contribute to the AI Dialogue by bringing complementary forms of expertise, implementation experience, and institutional perspective into a shared multilateral governance process. Governments and regulators can contribute legal, policy, and public sector governance experience, particularly in areas such as accountability, enforcement, rights protection, procedural fairness, explainability requirements, and oversight of automated decision-making systems. Technical communities and private sector actors can contribute insight into system design, deployment realities, interoperability challenges, and technical standards. Civil society, academic institutions, labour organisations, disability inclusion advocates, and affected communities can help surface governance gaps, lived impacts, and public accountability concerns that may not always be visible through technical or regulatory discussions alone. To ensure meaningful participation, the Dialogue should move beyond high-level principles and create structured engagement around operational governance challenges and implementation realities. This is particularly important because governance influence, technical capacity, and institutional readiness remain unevenly distributed across jurisdictions and stakeholder groups. A multi-layered structure may therefore be most effective. This could include: high-level plenary discussions on strategic governance priorities, technical and operational working groups, regional dialogue forums, cross-sector implementation sessions, and ongoing knowledge-sharing and review mechanisms between formal sessions. Practical implementation areas such as verification standards, auditability, enforcement coordination, institutional readiness, decision traceability, and accountability mechanisms should form part of these engagements. The Dialogue should also support meaningful participation from Global South contexts, not only as implementation environments, but as contributors to governance thinking and institutional practice. This includes recognising governance experiences emerging from different legal, developmental, and social contexts. Ultimately, the AI Dialogue will be most effective if participation is not only broad, but connected to practical governance implementation, institutional learning, and public legitimacy.

Several perspectives remain underrepresented in global AI governance discussions, particularly those emerging from Global South contexts, neurodiverse communities, and institutions operating within unequal developmental and governance conditions. Many existing governance frameworks are still shaped primarily through Euro-American regulatory, technical, and institutional perspectives. While these frameworks provide important foundations, there is a growing risk that governance approaches are adopted through regulatory replication rather than contextual adaptation. Governance models designed for highly resourced institutional environments may not translate seamlessly across different legal, social, economic, and administrative realities. Perspectives from the African continent and broader African diaspora are especially important because they reflect governance experiences shaped by institutional diversity, multilingual environments, uneven infrastructure, informal economies, and different social accountability traditions. These experiences may contribute valuable insights on institutional adaptability, relational accountability, and governance under constraint, particularly as AI governance becomes increasingly global and interconnected. Neurodiverse and neuroinclusive perspectives are also significantly underrepresented. As AI systems increasingly shape access to employment, education, healthcare, financial services, and public systems, governance discussions must consider whether affected individuals can meaningfully understand, challenge, and navigate AI-influenced decisions. This is particularly important in contexts where existing governance obligations already recognise rights relating to procedural fairness, explainability, and the ability to challenge automated decision-making outcomes. Systems that are technically compliant may still remain inaccessible, exclusionary, or difficult to interpret in practice. More inclusive governance will require moving beyond one-size-fits-all governance models toward approaches that are operationally adaptable, socially legitimate, and responsive to different institutional and human realities. This should include more regionally grounded participation structures, interdisciplinary engagement, and stronger inclusion of lived experience perspectives in governance development processes.

Innovative engagement formats for the AI Dialogue should move beyond traditional conference-style discussions and create spaces where governance experiences from different institutional, social, economic, and developmental contexts can meaningfully shape global AI governance conversations. One important approach would be regional implementation dialogues that allow jurisdictions to discuss how AI governance operates under different institutional realities, rather than assuming uniform levels of technological advancement, infrastructure, or regulatory capacity. This is particularly important in many developing and under-resourced contexts where AI adoption and institutional readiness are evolving at different speeds. The Dialogue could also benefit from comparative governance labs, implementation simulations, and institutional readiness exercises where participants examine how existing legal frameworks, governance systems, and accountability structures are already responding to AI-related challenges in practice. Innovation should not focus only on building entirely new frameworks, but also on repositioning and adapting existing governance systems to emerging AI environments. Multi-sector implementation forums involving regulators, technical communities, labour organisations, disability inclusion advocates, educators, civil society, and affected communities could also strengthen practical governance learning across sectors. To avoid governance approaches becoming overly concentrated around highly resourced regions, the Dialogue should support stronger participation from Global South and African diaspora perspectives, particularly where these communities navigate hybrid legal, social, and institutional realities across different governance environments. Interactive formats such as operational case studies, decision-accountability exercises, and cross-jurisdiction governance reviews could help shift discussions from high-level principles toward practical implementation challenges. Ultimately, the most effective engagement formats will be those that support mutual learning, contextual adaptation, and operationally realistic governance approaches across diverse global contexts.

Several existing policies, frameworks, and governance initiatives provide useful foundations for effective AI governance and could inform the work of the AI Dialogue. At the normative level, the UNESCO Recommendation on the Ethics of Artificial Intelligence and the OECD AI Principles establish important foundations for human rights, accountability, transparency, interoperability, and trustworthy AI governance across jurisdictions. At the operational level, the EU AI Act, the NIST AI Risk Management Framework, and ISO/IEC 42001 offer practical approaches to risk classification, governance processes, auditability, institutional oversight, and AI management systems. These frameworks are particularly valuable because they move beyond high-level principles toward implementation and organisational accountability. Regionally, the African Union Continental AI Strategy is significant because it recognises varying levels of digital readiness, institutional capability, and developmental context across African states. This suggests that effective governance cannot assume uniform technological advancement, infrastructure, or governance capacity across all societies. Emerging coordination mechanisms such as AI councils, regulatory cooperation platforms, and recent G7 discussions on trustworthy and human-centered AI also demonstrate growing recognition that AI governance requires institutional coordination and multilateral cooperation, not only standalone regulation. However, these approaches should not be treated as universally transferable without contextual adaptation. Governance models developed within highly resourced institutional environments may not always translate seamlessly across different legal, social, economic, and infrastructural realities. The added value of the AI Dialogue would therefore be its ability to connect these initiatives while supporting interoperability, institutional readiness, operational accountability, and more equitable participation in shaping global AI governance approaches.