WINAR Institute for AI Policy and Governance in Africa

Success is not another framework document. Success is when the outcomes of this Dialogue are actually implementable by a government regulatory body that was formed eighteen months ago, or a hospital in rural Kenya that has no compliance team. That means three concrete things. First, a shared accountability standard for high-risk AI deployments not principles, but a documented minimum: who authorized the system, under what conditions, who is liable when harm occurs. Simple, enforceable, universally applicable. Second, governance tools designed for low-resource institutional environments. Audit toolkits. Assessment templates. Practical instruments that organizations without legal teams can actually use. If the only people who can implement the outcomes of this Dialogue are already-resourced institutions in high-income countries, this Dialogue has failed. Third, a mechanism that gives developing country governments real power to interrogate AI systems deployed by international vendors within their borders. Not a recommendation. A mechanism. Success looks like Nairobi, Lagos, and Accra being able to use what comes out of Geneva. That is the test.

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

These four are not separate issues in African institutional contexts, they are the same problem described from different angles. You cannot have transparency without the capacity to demand it. You cannot build safe AI without accountability infrastructure that names who is responsible when something goes wrong. You cannot build capacity in isolation from governance standards that give that capacity meaning. And you cannot have any of it function across the continent without interoperability because Africa is 55 countries with 55 different regulatory timelines, and the governance tools need to work across that reality. From where I sit, advising organizations deploying AI in healthcare, education, financial services, and public administration across Africa, the most dangerous gap is not malicious intent. It is the absence of accountability infrastructure. Organizations are deploying AI systems they cannot interrogate, in contexts those systems were never validated for, with no documented chain of responsibility. That is a transparency failure, a safety failure, a capacity failure, and a governance interoperability failure all at once. These four priorities, addressed together, are the only way to close it.

Contextual validation as a deployment requirement. Every conversation about AI governance assumes the governance question is about how systems are built. In Africa, the more urgent question is about how systems are deployed in contexts they were never designed for, on populations whose data was never in the training set, in languages that were not considered, inside institutions that have no mechanism to push back. There is currently no international governance standard that requires an AI system to be validated for the specific context in which it will be deployed before that deployment happens. This gap is most acute in high-risk sectors, healthcare, education, public administration and most acute in low-resource institutional environments where the organization receiving the AI system has no power to demand documentation from the vendor providing it. This Dialogue should produce a contextual validation standard. Not a best practice. A baseline requirement for any high-risk AI deployment crossing borders or institutional contexts.

Kenya is currently debating the AI Bill 2026, the first comprehensive AI legislation proposed in East Africa. It is still in the Senate. That process alone is significant. But what I am watching closely is not whether the bill passes. It is whether, when it does, there will be any implementation infrastructure to back it up. Here is what the gap actually looks like on the ground. Organizations in Kenya and across Africa are deploying AI in healthcare, education, financial services, and public administration right now. Not planning to, doing it. And in the majority of those deployments, nobody can tell you who authorized the system, whether it was validated for the local context, or who is accountable when it causes harm. That accountability chain does not exist. Not because anyone is hiding it. Because nobody built it. The opportunity is real and urgent. Kenya's AI Bill creates a regulatory moment, a window where organizations are actively looking for guidance on what responsible AI looks like in practice. That window will not stay open indefinitely. The challenge is that the governance tools being offered to fill that gap are mostly imported. Frameworks designed for environments with mature regulatory bodies, large legal teams, and established compliance cultures. They do not translate to a district hospital, a government ministry, or an NGO operating across multiple languages and contexts. The most significant opportunity this Dialogue can create for Africa is governance instruments that are actually usable here, not adapted European compliance frameworks, but tools built for the institutional environments where AI's consequences are most acute and oversight capacity is most limited.

The AI Dialogue can do something no existing framework has managed to do: create a space where the countries most affected by AI governance failures have equal standing with the countries producing the most AI. That is not a small thing. Right now, international cooperation on AI governance largely means high-income countries setting standards and everyone else adapting to them. The OECD Principles, the EU AI Act, the G7 Hiroshima Process, these are serious efforts, but they were not built with African institutional realities at the table. The Dialogue can change that, but only if it is structured to produce outputs that developing countries can actually use, not just endorse. Concretely, the Dialogue can advance cooperation in three ways. First, by creating a shared baseline for accountability in high-risk AI deployments that works across different regulatory maturity levels, not a one-size-fits-all framework, but a minimum standard flexible enough to be implemented in Nairobi and Brussels alike. Second, by establishing a mutual recognition mechanism for national AI governance frameworks, so that countries building their own regulatory infrastructure are not starting from zero each time. Third, by creating a technical assistance pipeline, connecting countries with emerging regulatory frameworks to those with more developed ones, not as aid recipients but as governance partners. The Dialogue's value is not in producing another document. It is in being the room where the countries that have historically been governed by other people's AI decisions get to shape the rules for the first time.

Several initiatives have laid important groundwork that the Dialogue should connect with rather than duplicate. The UNESCO Recommendation on the Ethics of AI is the broadest multilateral consensus on AI governance principles currently in existence. The Dialogue should build directly on it specifically on its provisions around human oversight, transparency, and the rights of affected communities and focus on moving from principles to enforceable standards. The OECD AI Principles and the Global Partnership on AI have developed practical frameworks and monitoring tools. The Dialogue should draw on these while explicitly addressing the gap they leave: both were designed for OECD member country environments and need significant adaptation to function in developing country contexts. The African Union's work on the Continental AI Policy Framework and the AU Convention on Cyber Security and Personal Data Protection are critical foundations for Africa-specific governance. The Dialogue should formally recognize these as legitimate regional governance architectures not parallel processes to be harmonized away, but contributions to a genuinely pluralist global framework. The ITU's AI for Good platform and UNDP's work on AI governance capacity building in developing countries represent existing UN infrastructure the Dialogue should activate rather than rebuild. What the Dialogue adds that none of these have delivered is universality with teeth, a space where every country has standing, where outputs are designed to be implemented at different levels of regulatory maturity, and where the gap between principle and practice is treated as the central problem to solve, not a footnote. The added value is not new ideas. It is the political will and institutional authority to turn existing ideas into binding cooperation.

The Dialogue will only produce useful outcomes if the people closest to AI's consequences are shaping the conversation, not just observing it. Civil society and practitioner organizations should not be in the room to validate decisions already made by governments and the private sector. They should be in the room as equal contributors with named speaking roles, structured input mechanisms, and visible influence on outputs. For format, three recommendations. First, regional pre-consultations before each session, not just online surveys, but facilitated convenings where practitioners from Africa, Asia, Latin America, and other underrepresented regions develop consolidated positions that are formally tabled at the Dialogue. This ensures the Geneva room reflects input that was actually generated in Nairobi, Jakarta, and Bogotá. Second, a practitioner track running parallel to the government and private sector tracks specifically for people who are implementing AI governance on the ground in under-resourced environments. Their input is different from a policy researcher's and different from a tech company's. It needs its own space. Third, outcomes published in accessible formats, not just formal UN documents. If the outputs of this Dialogue cannot be read and used by a regulatory officer in an African country who does not have a team of lawyers, then the Dialogue has not done its job.

Three groups are consistently underrepresented and their absence distorts the conversation. First, African practitioners, not African governments, not African academics, but the people actually deploying and governing AI in African institutional contexts. Government delegations represent state interests. Practitioners represent implementation reality. These are not the same thing. The Dialogue needs explicit mechanisms to include practitioners from civil society and advisory organizations who are working inside the governance gap, not just describing it from outside. Second, communities directly affected by high-risk AI deployments, patients whose triage decisions are being made by AI systems, farmers whose crop advisory is algorithm-driven, loan applicants whose creditworthiness is determined by models they cannot see or contest. These communities are the reason AI governance matters. Their complete absence from global governance conversations is not an oversight. It is a structural failure. Participatory governance mechanisms, community consultation processes that feed into national positions are the only way to correct it. Third, women and girls in the Global South. AI systems are not gender-neutral. The governance conversations shaping those systems are overwhelmingly male and overwhelmingly from high-income countries. Dedicated mechanisms, fellowships, funded participation, structured speaking opportunities are needed, not aspirational language about inclusion.

The formats that produce the most honest and useful governance thinking are not the ones that look most impressive on a programme. Three formats worth building into the Dialogue First, structured case-based discussion, not presentations, not panels, but facilitated groups working through a real governance failure or deployment dilemma together. This is how practitioners actually think. It produces insight that keynote speeches never reach. Second, red team sessions, where a proposed governance standard or framework is actively challenged by practitioners from low-resource environments. If a framework cannot survive being stress-tested by someone trying to implement it in a district hospital in Kenya, it is not ready. Build that stress test into the Dialogue formally. Third, asynchronous regional input mechanisms that run between sessions, not just surveys, but structured deliberation processes that allow regions to develop and refine positions over time, not just submit written inputs once. The Dialogue meets in July 2026 and May 2027. That gap is an opportunity for ongoing structured engagement, not silence. The goal is a Dialogue that produces outputs shaped by the full range of people it is meant to serve, not a room of delegates producing documents that practitioners then try to adapt after the fact.

I want to share two approaches I have developed and deployed directly in African institutional contexts. The first is accountability mapping as a governance practice. Before any AI system is adopted by an organization, three questions must have documented answers: who authorized this system, under what conditions was it approved for this specific context, and who bears liability when harm occurs. This sounds simple. It is not standard practice. In my advisory work across African organizations, the absence of this accountability chain is the single most common governance failure I encounter. Making accountability mapping a baseline requirement, not a best practice is the most practical, lowest-barrier governance intervention available right now. The second is the AI Impact and Ethics Assessment framework I developed specifically for African organizational contexts. It covers six areas: fairness and inclusion, transparency and explainability, accountability and governance, privacy and security, societal and environmental impact, and monitoring and continuous improvement. What makes it different from existing frameworks is that it was designed for organizations without large compliance teams, practical, structured, and actionable without requiring legal expertise to implement. At the regional level, the AIGN Education Trust Label that I developed through the AI Governance Network is a promising model for sector-specific governance architecture. It creates a trust signal for AI tools in educational contexts, with accountability mapping built in as a core requirement. It is currently being piloted in African markets as a regional implementation of global governance principles adapted for local deployment realities. The common thread across all three is this: effective AI governance in under-resourced environments does not come from more principles. It comes from simple, structured, enforceable accountability tools that any organization can implement regardless of the size of their legal team.