Nasarawa State University Keffi, Nigeria

Success for the first Global Dialogue on AI Governance cannot be measured by the elegance of the framework it produces. It must be measured by whether that framework works for the contexts most likely to be harmed by ungoverned AI and least equipped to govern it themselves. Three outcomes would make this Dialogue genuinely successful. First, a governance architecture that addresses not just what AI systems output, but what they are permitted to access. My research experience digitalising records for a West African regional institution revealed that sensitive documents were not consistently protected by role-based access controls- meaning an AI system deployed in that environment could ingest confidential operational data and expose it through its outputs, with no malicious intent required. Current frameworks are almost silent on this. Success means the Dialogue establishes that AI access control is an AI safety issue, not merely an IT one. Second, capacity-building recognised as a governance obligation, not development assistance. If the Dialogue produces norms that only well-resourced states can implement, it has created a compliance illusion; not governance. Binding commitments to technical assistance for regional institutions in Africa and the Global South must be part of the core framework, not an annex. Third, meaningful inclusion of Global South researchers and practitioners in standard-setting not only as consultees in this process, but as permanent participants in the bodies that will interpret, update, and enforce whatever framework emerges. The hundreds of submissions gathered here are a starting point. The question is whether the voices behind them shape the outcome, or simply inform it. The global AI governance system we build now will be difficult to rebuild later. A Dialogue that centres those with the most to lose from getting it wrong will have succeeded.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight

Safe, secure and trustworthy AI is selected as the primary priority because safety cannot be assumed away. Field experience revealed that AI systems deployed in institutional environments without adequate access controls risk ingesting and inadvertently exposing sensitive operational data. This is a safety and security risk that current frameworks do not adequately address. Trustworthy AI must begin with governing what AI systems are permitted to access, not only what they are permitted to output. AI capacity-building is selected because the governance capacity gap is, in practice, the most immediate barrier to safe AI deployment across African institutions. Where institutions lack the technical expertise to audit AI systems, evaluate vendor claims, or contest automated decisions, governance frameworks however well-designed- cannot function. Capacity-building is not a complement to governance; it is a precondition for it. Transparency, accountability, and human oversight is selected because fragile institutional environments are precisely where accountability mechanisms are weakest and where AI-related harms are therefore most likely to go undetected and unaddressed. Vendor accountability, algorithmic auditability, and meaningful human oversight must be built into governance frameworks as binding obligations, particularly for deployments in low-capacity settings. Social, economic, ethical, cultural, linguistic and technical implications of AI is selected because governance frameworks designed without accounting for Africa's diverse institutional, linguistic, and social contexts will systematically fail the populations they are intended to protect. The implications of AI are not uniform across regions, and the Dialogue must produce frameworks sensitive to this heterogeneity.

The first is AI input risk in sensitive institutional environments specifically, the absence of governance standards for what data AI systems are permitted to access and ingest. Current frameworks concentrate almost entirely on AI outputs: what systems decide, recommend, or communicate. But in institutional settings where access controls are inconsistently enforced, the more immediate risk is on the input side. An AI system granted unrestricted access to an institutional document can expose confidential operational, financial, or conflict-sensitive information through its outputs, without any malicious design. This is simultaneously a cybersecurity issue, a data governance issue, and an AI safety issue- yet it falls between the gaps of all three thematic areas listed. Governing AI access to data must become an explicit component of the global framework. The second is the digitalization transition gap. Governance frameworks implicitly assume that the institutions deploying AI have already completed the transition from analogue to digital records management. In reality, millions of institutions across the Global South are currently mid-transition- operating with partially digitised, inconsistently formatted, and unevenly protected data. This transitional period is precisely when AI adoption risk is highest, because data quality, institutional readiness, and oversight capacity are all simultaneously incomplete. Yet no existing governance theme addresses the specific risks that arise during this transition. A truly inclusive global framework must account for institutions that are building their digital infrastructure and adopting AI at the same time because that is the lived reality for a significant portion of the world the Dialogue is meant to serve. Both issues are grounded in field observation, not theory, and both deserve dedicated attention in the governance framework that emerges from this process.

The governance gaps across the four selected thematic areas have direct consequences for West Africa's institutional landscape. The most significant challenge is the deployment-governance mismatch. AI tools are being adopted by regional institutions and governments at a pace that far outstrips the frameworks meant to manage them. Institutions are procuring AI-assisted systems for financial management, logistics, and record-keeping without the technical capacity to audit those systems, without harmonised legal frameworks governing cross-border data, and without access control standards adequate to the sensitivity of the data involved. Governance gaps here are not abstract, they translate directly into institutional vulnerability. A second challenge is vendor asymmetry. West African institutions are active markets for AI vendors who frequently operate with minimal transparency about how their systems work or what happens when they fail. Without binding international standards for vendor accountability in low-capacity environments, institutions bear all the risk while vendors bear almost none. The opportunity lies in the regional governance moment. West Africa has existing institutional architecture with mandates covering peace, security, and economic integration that could anchor continent-relevant AI governance. The challenge is that these institutions need capacity support before they can meaningfully implement global norms- let alone help shape them. Finally, for emerging researchers in cybersecurity and digital forensics across the region, the governance gap is also an opening. African expertise in these fields is directly relevant to AI safety problems the global community has not yet adequately theorised. Investing in this research pipeline is both a capacity-building imperative and a long-term governance dividend.

The AI Dialogue arrives at a moment when the international community faces a fundamental choice: whether AI governance reproduces existing global inequalities, or whether it builds something more equitable from the outset. The most important contribution the Dialogue can make is to establish cooperation as a structural feature of governance, not a diplomatic courtesy. This means moving beyond declarations of intent toward binding mechanisms for capacity transfer, shared technical standards, vendor accountability across borders, and meaningful participation of Global South institutions in bodies that will enforce whatever framework emerges. The Dialogue is also well positioned to bridge the fragmentation of existing governance efforts. The EU AI Act, OECD AI Principles, the African Union's emerging frameworks, and various national approaches currently operate in parallel with limited coordination. A cooperative outcome would establish interoperability between these frameworks without flattening the contextual differences that make regional approaches necessary. Most critically, the Dialogue can make legitimate the voices of practitioners and researchers from underrepresented regions in ways that technical standard-setting bodies historically have not. Field-level insights about data readiness, access control risks, and institutional fragility are governance-relevant knowledge not local colour. International cooperation must be informed by these perspectives, not only by the priorities of the most technologically advanced states. Cooperation that flows in one direction is not cooperation- it is instruction. The Dialogue has the mandate and the moment to model something better.

Several existing initiatives provide foundations the AI Dialogue should build upon rather than duplicate. The African Union's Continental AI Strategy represents the most significant regional effort to develop Africa-owned governance principles. The Dialogue should formally connect with this framework, ensuring global norms do not marginalise the contextual priorities African states have already identified collectively. The ITU's AI for Good platform has established valuable convening infrastructure. The Dialogue should leverage this network but shift emphasis from showcasing AI applications toward governing them particularly in fragile and transitional institutional contexts. The OECD AI Principles and Global Partnership on AI (GPAI) have produced substantive groundwork but suffer from significant underrepresentation of African and Global South voices. The Dialogue's added value here is legitimacy: its UN mandate allows it to make inclusion a binding condition rather than an aspiration. Regional economic communities already have institutional mandates covering cross-border data, digital infrastructure, and security cooperation. The Dialogue should recognise these bodies as legitimate governance actors and invest in their capacity to implement and adapt global frameworks to regional realities. The clearest added value the Dialogue brings is universality. No existing initiative spans the full UN membership. This reach creates a unique opportunity not to impose a single framework, but to build connective tissue between regional, national, and sectoral approaches that currently operate in isolation. That connective role, executed with genuine inclusivity, is something no other mechanism can replicate.

Meaningful contribution to the AI Dialogue requires more than open registration and submission portals. The format and structure must be deliberately designed to surface knowledge that traditional diplomatic forums routinely miss. For academia and research, the Dialogue should create dedicated channels for field-based researchers not only those affiliated with well-funded institutions in high-income countries. Postgraduate and early-career researchers from the Global South frequently possess the most grounded insights into how AI is actually being adopted and what governance gaps look like in practice. Reserved participation for regional research communities would ensure these perspectives shape outcomes rather than merely populate consultation records. For civil society, the Dialogue should invest in facilitating regional convenings particularly in Africa, Southeast Asia, and Latin America where affected communities can develop collective positions before the global forum, not after it. For governments, the Dialogue should distinguish between states with advanced regulatory capacity and those still building foundational digital governance frameworks. Tiered participation structures, paired with technical assistance, would allow all member states to engage substantively rather than symbolically. For industry, participation should be welcomed but structured to prevent regulatory capture. Vendor contributions should be transparently disclosed and balanced against the voices of communities that bear the consequences of deployment decisions. On format, the Dialogue should avoid producing a high-level declaration as its primary output. Actionable working group recommendations, regional implementation roadmaps, and binding capacity commitments would constitute a far more durable legacy. The measure of inclusive design is simple: can those with the least institutional power still meaningfully participate?

Several voices remain systematically underrepresented in global AI governance discussions, and their absence weakens the frameworks being produced. Postgraduate and early-career researchers from the Global South possess field-level knowledge about how AI is being adopted in fragile, transitional, and under-resourced institutional contexts: knowledge that is directly governance-relevant but rarely reaches standard-setting bodies. Dedicated fellowship programmes and reserved participation in working groups would begin to address this gap. Regional institutions and economic communities in Africa have existing mandates covering digital infrastructure, cross-border data, and security cooperation, yet are largely absent from the forums where global AI norms are being shaped. Formal observer and contributor status for these bodies within UN-led governance processes would strengthen both inclusivity and implementation capacity. Practitioners working at the intersection of digitalization and AI adoption including those managing institutional record systems, data infrastructure projects, and digital transition programmes across the Global South hold insights about real deployment conditions that theorists and policymakers lack. Structured practitioner tracks within the Dialogue would surface this knowledge systematically. Inclusion cannot be achieved through open portals alone. It requires active investment: in translation, in regional convenings, in financial support for participation, and in governance structures that give underrepresented voices decision-making power, not merely the opportunity to be heard.

Traditional conference formats- plenary speeches, panel discussions, and high-level declarations are poorly suited to generating the practical, grounded dialogue that effective AI governance requires. The following formats would foster more meaningful engagement. Regional practitioner roundtables held prior to the main Dialogue would allow researchers, institutional actors, and civil society from underrepresented regions to develop consolidated positions before arriving at the global forum. This shifts participation from reactive to substantive. Case-based working sessions where participants present real deployment scenarios including failures and governance gaps encountered in the field would ground abstract policy discussions in operational reality. Anonymised case studies from fragile and transitional institutional contexts would be particularly valuable. Structured dialogue between researchers and policymakers in small, facilitated groups would create space for the kind of frank exchange that large plenary formats discourage. Early-career researchers and postgraduate voices should be explicitly included in these sessions, not siloed into separate youth tracks. Open synthesis sessions at the end of each thematic area, where a diverse rapporteur group consolidates key insights across stakeholder categories, would ensure that field-level knowledge informs final recommendations rather than disappearing into background documentation. The goal is a Dialogue that generates actionable knowledge and not one that merely documents the positions participants arrived with.

Several existing policies and practices offer concrete models worth building upon. The EU AI Act's risk-tiered approach- categorising AI systems by risk level and applying proportionate requirements offers a useful structural model. However, its implementation burden assumes regulatory capacity that most African institutions do not yet possess. An adapted version that incorporates institutional readiness as a tiering criterion, not just application risk, would be more globally applicable. Rwanda's National AI Policy demonstrates that Global South governments can lead on AI governance when given adequate resources and policy space. It offers a replicable model for nationally owned frameworks that balance innovation with accountability. The NIST AI Risk Management Framework provides practical, flexible guidance for identifying and managing AI-related risks. Its adaptable structure makes it more suitable for diverse institutional contexts than prescriptive regulatory models. At the practitioner level, cybersecurity access control principles: least privilege, need-to-know, sensitivity classification offer an immediately applicable model for governing what data AI systems are permitted to ingest. Extending these established frameworks to AI data access governance requires no new invention, only intelligent adaptation. Finally, participatory AI assessment approaches being piloted by civil society organisations in Africa where affected communities evaluate AI systems before and during deployment offer a grassroots accountability model that complements top-down regulation. Effective AI governance does not always require new frameworks. Often it requires the intelligent extension of what already works.