University of Pretoria

A successful first Global Dialogue on AI Governance, scheduled for 2026, would be defined by its ability to transition from high-level principles to actionable, inclusive multilateralism. For this inaugural session to be considered a success, I believe it must achieve the following three outcomes: 1. Meaningful Inclusivity and Representation A primary goal is bridging the "AI divide." Success means ensuring that the 118+ countries currently excluded from global AI discussions have a functional seat at the table. This requires not just their presence, but a concrete commitment to capacity building, including voluntary financing mechanisms that allow developing nations to participate in both the governance and the economic benefits of AI. 2. Interoperability of Governance Frameworks With AI regulations currently fragmented (e.g., EU AI Act, US Executive Orders, China's regulations), success would involve establishing a "common language." The Dialogue should produce a roadmap for interoperability, ensuring that different national and regional rules align on core safety and human rights standards to prevent a fractured global digital ecosystem. 3. Integration of Science and Policy The Dialogue is uniquely paired with the UN's Independent International Scientific Panel on AI. A successful outcome would be the formal adoption of an evidence-based risk assessment framework. This would ensure that future policy decisions are grounded in the Panel's scientific findings rather than political or corporate interests, particularly regarding existential risks and algorithmic bias. Ultimately, the Dialogue succeeds if it establishes the UN as the permanent, stable home for AI coordination—transforming the Global Digital Compact from a vision into a living, responsive governance architecture.

• Safe, secure and trustworthy AI
• AI capacity-building

To achieve a truly cohesive global framework, the Dialogue must prioritize the interoperability of governance approaches, ensuring that diverse national and regional regulations-such as the EU AI Act, US Executive Orders, and various frameworks in the Global South-can function together seamlessly. By focusing on safe, secure, and trustworthy AI through the lenses of transparency, accountability, and human oversight, the international community can create a "common language" for risk management that prevents a fractured digital landscape. This interoperability, supported by robust AI capacity building, ensures that all nations, regardless of their developmental stage, can adopt and align with these global standards, ultimately fostering a secure, inclusive, and innovation-friendly environment that protects human rights across borders.

While the current themes cover the primary pillars of AI governance, several critical cross-cutting and emerging issues deserve more explicit attention: 1. Environmental Sustainability and Resource Impact: The environmental cost of AI is often sidelined. The Dialogue must address the massive energy consumption of data centers and the heavy water usage required for cooling. A governance framework is incomplete without standards for "Green AI" and transparency regarding the carbon footprint of training large-scale models. 2. Labor Displacement and the Future of Work: While "capacity building" focuses on skills, there is an urgent need for a global conversation on the socio-economic disruption caused by AI automation. This includes cross-border social safety nets, the protection of workers' rights in the "gig" and data-labeling economies, and the potential for increased global wealth inequality. 3. The "Data Commons" and Cultural Sovereignty: Current AI models are often trained on datasets that lack linguistic and cultural diversity, leading to "digital colonization." An emerging priority is the creation of a global "Data Commons" that protects the intellectual property and cultural heritage of indigenous and non-Western populations from being exploited or erased by AI training processes. 4. AI in Conflict and Dual-Use Proliferation: While "safety and security" are mentioned, the specific intersection of AI with autonomous weaponry and its potential to lower the threshold for conflict is a distinct, high-stakes issue. Establishing clear international norms on the "dual-use" nature of open-source models is vital to prevent non-state actors from accessing dangerous capabilities. Addressing these issues ensures that AI governance is not only technically sound but also ecologically responsible and socially just.

In South Africa, the intersection of AI governance gaps and rapid technological advancement presents a complex landscape of risk and potential. Significant Challenges The most pressing challenge is the widening digital and "AI divide." Without robust AI capacity building, South Africa risks becoming a mere consumer of foreign technologies that do not account for local linguistic diversity or socio-economic contexts. The lack of a comprehensive, AI-specific national regulatory framework creates a governance gap in transparency and accountability, particularly regarding algorithmic bias. In a country with a history of systemic inequality, unregulated AI in sectors like credit scoring, hiring, and law enforcement could inadvertently entrench historical prejudices. Furthermore, the high energy demands of AI infrastructure pose a significant challenge to an already strained national power grid, highlighting the urgent need for Green AI standards. Significant Opportunities Conversely, AI offers a transformative opportunity to leapfrog traditional developmental hurdles. By prioritizing safe, secure, and trustworthy AI, South Africa can lead the African continent in creating a "pro-innovation, ethical-first" ecosystem. There is immense potential for AI for Sustainable Development, particularly in optimizing public healthcare delivery, improving agricultural yields for small-scale farmers through predictive analytics, and enhancing educational outcomes in under-resourced schools. The move toward interoperability of governance approaches provides South Africa with the opportunity to align its Protection of Personal Information Act (POPIA) with global standards, making the country an attractive and secure hub for ethical AI investment and data processing. By addressing these gaps, South Africa can ensure that AI serves as a tool for social redress and economic inclusion rather than a driver of further disparity.

The Global Dialogue on AI Governance can serve as the definitive multilateral platform to transition from fragmented national policies toward a cohesive international framework. Its primary role is to act as a normative anchor, providing a stable, inclusive space within the United Nations where all member states—regardless of their technological maturity—can shape the rules of the digital age. First, the Dialogue can drive regulatory interoperability. By establishing a shared vocabulary and baseline safety standards, it can prevent a "race to the bottom" or a fractured global market. This alignment reduces compliance burdens for innovators while ensuring that core principles like transparency, accountability, and human oversight are non-negotiable across borders. Second, it acts as a critical bridge for the AI divide. Through the formalization of capacity-building mechanisms, the Dialogue can facilitate the transfer of technical expertise, sustainable infrastructure, and governance toolkits to the Global South. This ensures that international cooperation is not merely symbolic but involves the active participation of developing nations in the global AI value chain. Third, the Dialogue provides a unique interface between science and policy. By integrating the findings of the Independent International Scientific Panel on AI, it ensures that global cooperation is grounded in objective, empirical risk assessments rather than political posturing. This evidence-based approach is essential for tackling cross-border challenges such as existential risks, algorithmic bias, and the environmental impact of large-scale AI. Ultimately, the Dialogue's most significant role is to institutionalize inclusive multilateralism, ensuring that AI governance is not dictated by a small group of tech-dominant nations, but is instead a collective effort to align artificial intelligence with the UN Charter and the Sustainable Development Goals.

The AI Dialogue should avoid duplicating existing efforts and instead serve as the connective tissue between fragmented initiatives. It must build upon the foundational work of the OECD AI Principles and the UNESCO Recommendation on the Ethics of AI, which have already established broad ethical consensus. Furthermore, it should integrate the technical safety standards emerging from the HIROSHIMA AI Process and the Bletchley Declaration, as well as the regulatory frameworks like the EU AI Act and the African Union's AI Strategy. The added value of the AI Dialogue lies in three specific areas: 1. Universal Legitimacy and Inclusivity: Unlike the G7 or OECD, the UN-led AI Dialogue provides a universal platform where the Global South often excluded from high-level tech diplomac has an equal voice. This ensures that governance standards are not just technically sound but globally representative and culturally diverse. 2. Institutionalized Science-Policy Linkage: By formally connecting with the Independent International Scientific Panel on AI, the Dialogue can translate complex, evolving scientific data into actionable multilateral policy. This provides a "single source of truth" for risk assessment that individual national initiatives currently lack. 3. A Centralized Clearinghouse for Capacity Building: While many partnerships (like the Global Partnership on AI -and GPAI) focus on innovation, the AI Dialogue can act as a global coordinator for resource distribution. It can match the surplus of technical expertise and infrastructure in developed nations with the specific developmental needs of the Global South, directly linking AI governance to the Sustainable Development Goals (SDGs). Ultimately, the Dialogue transforms a "patchwork" of voluntary guidelines into a formalized, global architecture, ensuring that international cooperation on AI is coherent, enforceable, and inclusive.

To be effective, the AI Dialogue must transition from a state-centric model to a multi-stakeholder architecture that reflects the borderless nature of AI. Contribution and structure should be defined by the following recommendations: Stakeholder Contributions • Private Sector & Technical Community: Must move beyond "self-regulation" to provide technical transparency and data for risk assessments. Developers should contribute to a global "compute and expertise bank" for capacity building in the Global South. • Civil Society & Academia: Act as the ethical conscience of the Dialogue, ensuring that human rights, gender equality, and labor protections are at the forefront. Academic institutions should drive the Independent Scientific Panel to ensure evidence-based policy. • Regional Organizations: (e.g., African Union, EU) Should serve as aggregators, harmonizing regional positions to streamline global negotiations. Format and Structure 1. Hybrid "Track" System: The Dialogue should operate via two parallel tracks: a Political Track for high-level diplomatic agreements and a Technical/Scientific Track for real-time risk monitoring and standard-setting. 2. Regional Consultative Hubs: To ensure inclusivity, the Dialogue should hold regional preparatory sessions. This prevents the main forum from being dominated by a few tech-heavy nations and allows local contexts (e.g., linguistic diversity, infrastructure gaps) to inform global rules. 3. An Interactive "Living" Repository: Instead of static annual reports, the structure should include a digital clearinghouse for AI governance toolkits, best practices, and a "matchmaking" platform for capacity-building resources. 4. Open Public Consultations: Utilizing AI-enabled platforms to gather global citizen input ensures that the "human-centric" goal of the Dialogue is grounded in the lived experiences of those most affected by AI deployment.

The most significant gap in global AI governance is the absence of Global South voices, specifically from Africa, Southeast Asia, and Latin America, which are often treated as data sources or testing grounds rather than architects of policy. Furthermore, Indigenous communities, youth, and small-to-medium enterprises (SMEs) remain largely marginalized in a landscape dominated by "Big Tech" and a few powerful nation-states. To ensure a truly inclusive AI Dialogue, we must move beyond symbolic presence toward structural inclusion: 1. Linguistic and Cultural Diversity: Current AI models and governance frameworks are heavily Western-centric. Inclusion requires the active participation of linguistic experts and cultural historians from underrepresented regions to prevent "digital colonization" and ensure AI systems respect diverse value systems and protect indigenous data sovereignty. 2. Institutionalized Youth Participation: Since the long-term impacts of AI—from labor displacement to existential risks—will most affect the next generation, the Dialogue should include a formal Youth Advisory Body with a mandate to review and comment on proposed policies. 3. Financial and Technical Subsidies: Participation is often limited by resources. The UN should establish a Travel and Technical Support Fund to ensure that civil society organizations and academic researchers from developing nations can attend sessions and contribute high-quality, localized research. 4. Regional Hubs for Grassroots Input: Instead of centralized meetings in Geneva or New York, the Dialogue should utilize regional consultative forums. This allows for the inclusion of local labor unions, teachers, and healthcare workers whose lives are being transformed by AI but who lack the platform to influence global standards. By prioritizing these perspectives, the AI Dialogue can transition from an elite diplomatic exercise to a globally legitimate framework that protects the rights and interests of all humanity.

To move beyond traditional, static diplomatic plenary sessions, the AI Dialogue should adopt dynamic formats that mirror the agility of the technology it aims to govern. Here are four innovative engagement formats: 1. The "Red-Teaming" Policy Hackathon Instead of just debating text, stakeholders (policymakers, hackers, and civil society) should participate in policy simulations. Participants would be presented with a hypothetical AI crisis—such as a cross-border deepfake campaign or an algorithmic collapse in financial markets—and tasked with "stress-testing" proposed governance frameworks to identify gaps in real-time. 2. Multi-Stakeholder "Fireside" Deliberative Polling Utilizing AI-enabled deliberation platforms, the Dialogue can aggregate real-time input from thousands of global citizens simultaneously. These "digital town halls" would allow underrepresented communities to vote on priority issues, with the results immediately integrated into the high-level diplomatic track, ensuring that "human-centric" is not just a buzzword but a data-driven mandate. 3. The "Reverse Pitch" for Capacity Building Inverting the traditional aid model, developing nations and regional bodies would "pitch" their specific governance and infrastructure needs to a panel of "investors" (tech companies, developed states, and NGOs). This creates a marketplace for cooperation, where commitments to cloud credits, expert training, or regulatory toolkits are made transparently and matched to local demand. 4. Interactive "Living" Evidence Labs The Independent International Scientific Panel on AI should host "Evidence Labs"—interactive exhibitions where policymakers can interact with AI models to visualize the impacts of different regulatory choices. Using data visualization to show how a specific policy might affect carbon emissions or labor markets in different regions makes abstract governance tangible and evidence-based. By shifting from monologue to co-creation, these formats ensure the Dialogue remains responsive to the rapid pace of AI evolution.

Effective AI governance is currently being shaped by a move from high-level principles to operational tools and inclusive regional strategies. Below are concrete examples of policies, platforms, and approaches that offer solutions to contemporary challenges: 1. Regulatory Policies: The "Risk-Based" Approach • The EU AI Act (2025): This is the most comprehensive "hard law" example, categorizing AI systems by risk level (Unacceptable, High, Limited, Minimal). It mandates strict transparency and human oversight for high-risk applications, such as those used in critical infrastructure or law enforcement, providing a clear legal blueprint for accountability. • Brazil's AI Strategy: Brazil has integrated its national AI strategy with public hearings, emphasizing a multi-stakeholder committee that includes academia and civil society to ensure that liability frameworks are not just technically sound but socially representative. 2. Specialized Platforms: Ensuring Transparency and Accountability • Responsible AI Toolkits (e.g., Azure Machine Learning, IBM Watson OpenScale): These platforms provide automated auditing for algorithmic bias and "drift." They allow organizations to monitor models in real-time, providing "explainability" features that show which data points influenced a specific decision, such as a loan rejection.