African countries are at an important juncture in AI governance. Several national AI strategies have been adopted across the region, and in 2024 the African Union endorsed its first Continental AI Strategy. Yet, the majority of countries still do not have national AI strategies, and those that have adopted them face common challenges in operationalising policy objectives and translating them into effective governance and implementation. This case study draws on insights from a workshop held in Nairobi, Kenya, which convened senior government officials from 12 African countries. The workshop was co-organised by the OECD, the Agence Française de Développement and the United Kingdom’s Foreign, Commonwealth & Development Office. This case study identifies emerging patterns, shared constraints, and practical approaches to AI governance across three thematic areas: agriculture, locally developed language models, and regulatory frameworks for AI. These insights will contribute to the ongoing development of the OECD AI Policy Toolkit, which will support economies in harnessing AI benefits.
AI governance in Africa
Abstract
Countries in Africa increasingly view AI as a tool to support economic development, public service delivery and regional integration. A growing number of governments have adopted national AI strategies, and continental commitments have strengthened through the African Union (AU)’s AI Strategy. Translating this momentum into effective governance, however, requires addressing structural constraints that differ from those in higher‑income economies, including infrastructure gaps, institutional capacity limitations and data governance challenges. These factors also shape which governance frameworks are feasible in practice. In this context, AI governance involves not only designing regulatory frameworks but also strengthening the technical, institutional, and human systems that enable their effective implementation.
The OECD AI Policy Toolkit under development aims to address these challenges by providing practical, context-sensitive guidance to support the design and implementation of AI policies aligned with the OECD AI Principles. This case study draws on findings from a workshop held in Nairobi, Kenya (10–11 March 2026), organised with the technical and financial contributions of the Agence Française de Développement and the Foreign, Commonwealth & Development Office of the United Kingdom, and co-hosted by Kenya’s Ministry of Information, Communications and the Digital Economy. The multistakeholder workshop convened government representatives and experts from 12 African countries and included a roundtable on shared challenges and panel sessions on agriculture, language models and regulation.
The findings are organised around insights from the roundtable on challenges and solutions, and the themes from the panel discussions: AI in agriculture, locally developed language models, and regulatory frameworks. These themes are mutually reinforcing, as AI in agriculture depends on locally relevant data and language models, and regulation shapes what is possible and how solutions can be responsibly deployed.
AI policy activity across African countries has accelerated since 2020, with several countries having adopted national AI strategies (Table 1). Common priorities include economic diversification, skills development and domestic technological capacity, with agriculture, health, education and public administration frequently identified as priority sectors for AI applications in Africa. At continental level, the AU’s Continental AI Strategy, adopted in 2024, establishes an overarching framework, and calls for the development of an African Charter on Trustworthy AI.
Note: Based on information available as of April 2026; information on Togo and Uganda provided at the workshop.
Shared implementation challenges
A set of interconnected challenges are shaping AI governance across the region:
Infrastructure, compute, and connectivity: Limited access to digital infrastructure, affordable connectivity and advanced compute capacity remains one of the most consistently cited constraints to develop, deploy, and sustain AI systems. Beyond financing, access to hardware such as GPUs is affected by global market concentration and geopolitical factors. Data centre development also depends on broader enabling conditions, including energy and water resources.
Sustainability of AI systems: Ongoing monitoring, maintenance and retraining require stable funding and technical capacity that are often lacking beyond pilot phases. Environmental sustainability was also raised as a concern, given the energy and water demands associated with AI.
Human capital, AI literacy, and brain drain: Skills shortages affect both technical and regulatory capacity and are compounded by talent outflows. AI literacy for citizens and public officials is best built by integrating AI concepts across education systems, rather than limiting them to specialist training.
Data governance and quality: Weak data quality and governance constrain AI system reliability and trust. Inconsistent data standards, limited interoperability, incomplete data classification and uneven enforcement of data protection frameworks are among the key barriers. Investments in interoperable data infrastructure are foundational and generate benefits across multiple AI use cases and sectors.
Financing and political buy-in: Implementing AI strategies requires sustained public investment alongside private and international finance. Fiscal constraints, declining official development assistance, and competing spending priorities make this difficult, particularly as AI benefits materialise over longer time horizons.
Governance capacity: Translating AI strategies into operational governance requires clear institutional mandates, inter‑ministerial co‑ordination and monitoring mechanisms that are still emerging in most countries.
Regulatory frameworks and institutional co-ordination: Designing proportionate AI governance frameworks is challenging when AI technologies evolve rapidly and are largely developed abroad. Most countries have institutions with relevant mandates, including data protection authorities, telecoms regulators, sectoral bodies, but technical capacity to assess AI‑specific risks remains limited. AI governance functions are distributed across ministries, with inter-institutional co-ordination mechanisms being at an early stage. Strengthening co-ordination mechanisms across institutions regional harmonisation are priorities, while recognising that different national contexts may produce different regulatory pathways.
A range of initiatives are already underway across the continent to address these challenges (Table 2).
AI governance and development outcomes in Africa: spotlight on AI in agriculture
Agriculture accounts for a substantial share of employment and economic output across much of the region - employing roughly 45–60% of the workforce and contributing around 15–20% of GDP across the continent -, yet productivity remains significantly below global averages, with output per worker several times lower than in advanced economies. AI applications, ranging from crop monitoring and pest detection to yield forecasting and advisory services, offer important opportunities to address persistent productivity gaps. Successful deployment does not primarily depend on frontier models, but on the availability of high-quality data, the effectiveness of distribution channels, the level of trust among users, and the affordability of solutions:
The foundational role of data. Soil maps, weather and climate information, market data, and agronomic response models form the essential base layer for AI in agriculture. Without these inputs, even the most advanced models cannot generate reliable, locally relevant recommendations. Such datasets that are foundational to a specific sectoral implementation are increasingly understood as digital public infrastructure and require sustained investment, standardisation, and governance.
The importance of distribution mechanisms. Farmers are unlikely to adopt new tools simply because they are available; uptake depends on whether advice is delivered through trusted channels, such as extension services, cooperatives, or familiar digital platforms. In this regard, human relationships remain central, and rather than replacing extension systems, AI is seen as a means of strengthening and scaling them.
Trust as a critical condition for adoption. Trust is built incrementally, through repeated demonstrations of value, rather than through one-off interactions. Farmers are more likely to adopt AI-supported practices when they can clearly observe tangible economic benefits, such as improved yields or reduced input costs. This highlights the importance of designing AI systems that align closely with farmers’ decision-making cycles and economic realities.
Significant barriers to scaling agricultural AI remain. Data scarcity and poor data readiness were identified as major limitations, with much of the information required for localisation fragmented, undigitised, or not available in AI-ready formats. Even where data exists, weak data-sharing practices often limit its use, as institutions may be reluctant or unable to make datasets accessible. Lack of interoperability presents an additional constraint. Legacy systems are often difficult to integrate, and multiple actors are developing solutions in parallel without common standards, leading to fragmentation. Moreover, smallholder farmers have very low willingness and often ability to pay directly for AI tools, such as AI advisory services. This raises questions about sustainable business models and the need for public or blended financing approaches. Language and accessibility barriers further complicate deployment, as advisory services must function across diverse linguistic contexts and varying levels of literacy and connectivity.
Taken together, these insights suggest that the scaling of agricultural AI in African contexts depends less on technological breakthroughs than on the development of robust, inclusive, and well-governed ecosystems that connect data, institutions, and end users effectively. The above-mentioned data quality and trust challenges are reflected at another scale in the debate on language models: the same structural barriers that limit AI uptake in agriculture, such as fragmented data, limited local representation, and uncertain institutional frameworks, also shape how African countries approach the development and governance of language models.
Locally developed Small Language Models and Large Language Models: opportunities and trade-offs
Language models represent a critical frontier for AI-enabled development in Africa, with the potential to bridge longstanding communication and service-delivery gaps at scale. In agriculture, language models can power AI advisory services that deliver crop recommendations and market information in local languages via mobile phones. In health, they can support community health workers with diagnostic guidance. In public administration, they enable automated document processing and multilingual interfaces that make government more accessible. In education, they underpin adaptive learning tools.
However, their effectiveness depends on whether the underlying models are trained on sufficient, high-quality data in African languages and calibrated to local contexts. African languages and contexts remain significantly underrepresented in the training data of most commercially available AI systems, affecting the accuracy, relevance, and reliability of these systems in African settings. This has prompted debate between “buy” approaches on the one hand, i.e. relying on commercial models for immediate access to advanced capabilities, but with concerns around cost, accuracy, dependency, and data sovereignty, and “build” strategies on the other, which offer greater control and alignment with local needs but require substantial upfront investment, highly specialised talent, and sustained operational funding. Country experiences illustrate a range of strategic approaches to developing language model capabilities in African contexts:
In Egypt, the development of Karnak, a national Arabic large language model, reflects efforts to align AI systems with local linguistic specificities and public-sector needs. Positioned as a form of “sovereign digital infrastructure”, Karnak is conceived both as a government tool and as an open-weight platform supporting local innovation ecosystems. This approach emphasises frugal AI principles, lifecycle risk management, and rigorous benchmarking. Karnak has been deployed in public administration contexts, including to support services in Arabic and to process government documents.
Nigeria’s development of Atlas, a multilingual model covering the three major national languages and Nigerian-accented English, illustrates a complementary strategy. While pursuing domestic model development, this approach recognises that training frontier-scale models is often prohibitively costly and may not represent an optimal allocation of resources. Instead, the country places emphasis on high-quality data curation, the development of domain-specific models for priority sectors such as agriculture, health, law, and public administration, and the creation of application ecosystems that translate AI capabilities into tangible economic value.
Complementing these national approaches, the Masakhane research community – supported by the UK alongside their AI for Development partners - demonstrates the potential of collaborative, pan-African models of innovation. Since 2018, this grassroots initiative has brought together researchers, linguists, and practitioners to develop open benchmarks covering over 100 African languages, improve global evaluation frameworks by identifying biases and gaps, and produce open-source datasets across tasks such as translation, classification, speech recognition, and hate-speech detection. This model emphasises resource-efficient language models, participatory data governance, sharing benefits with language communities, and the development of standards to reduce fragmentation and strengthen inclusiveness.
National approaches to language model development can be further enhanced and sustained through regional collaboration, including shared investment in compute infrastructure, common standards and benchmarks, and data governance frameworks enabling trusted cross-border data sharing. The policy challenge is less about ‘build or buy’ than about positioning within the global AI value chain, as neither approach, taken in isolation, appears viable for most countries. A more effective pathway lies in a strategic middle ground that combines locally relevant data, targeted model development, and strong institutional frameworks, with evidence suggesting that innovation and value creation are often concentrated at the application layer, rather than in the development of frontier models.
A central governance question is how African countries can design effective, rights-centred approaches to AI regulation that also support innovation. This requires balancing the opportunities of AI-driven development with the need to protect fundamental rights, ensure accountability, and sustain public trust. Evidence from cross-country experience suggests that:
AI governance does not require the immediate adoption of comprehensive AI-specific legislation. Most African countries have relevant legal and regulatory frameworks, including data protection, consumer protection, cybersecurity, and sectoral regulations, that provide a foundation for AI oversight. Strengthening and adapting these existing regimes is often a more feasible path to AI governance than introducing entirely new legislation. This approach implies treating AI as a cross-cutting technology, with existing regulators continuing their mandates while progressively building the technical capacity required to oversee AI-enabled systems.
The trade-off between regulation and innovation is a false dichotomy. Predictable and principled governance frameworks can reduce uncertainty and support more sustainable innovation. Conversely, regulatory gaps may increase risks and undermine trust, as firms adapt to unregulated spaces. In this sense, well-designed governance frameworks function as an enabling condition for responsible innovation rather than the opposite.
Risk-based approaches are increasingly seen as a pragmatic basis for regulation. Rather than focusing on AI technologies in the abstract, governance frameworks can be structured around use cases and their potential impacts. Under this approach, regulatory requirements scale with risk: high-impact applications, such as those in healthcare, justice, or biometric identification, require stricter oversight, while lower-risk uses can operate under lighter oversight regimes. This requires regulators to define acceptable levels of risk explicitly, in order to avoid overly precautionary approaches that may constrain beneficial uses.
Regulatory experimentation as an important governance tool. Mechanisms such as regulatory sandboxes enable controlled testing of AI systems under supervision, reducing uncertainty for both regulators and innovators. These approaches also support institutional learning and capacity building, particularly in environments characterised by rapid technological change and limited regulatory resources. Although not yet operational, Senegal’s approach considers a sandbox as part of their policy on Startups.
The governance of AI systems developed and operated outside national jurisdictions introduces additional complexity. Available policy levers remain limited but include fiscal measures, such as taxation of locally generated revenues and principle-based conditions for market access. While partial, these approaches provide entry points for addressing cross-border governance challenges.
Effective AI regulation in these contexts focuses on the governance of impacts rather than of technology itself, using a balanced mix of principles, risk-based frameworks, experimentation, sectoral oversight, and continuous institutional learning.
Building on existing legal and institutional frameworks. Before drafting new AI legislation, policymakers should assess existing data protection, consumer protection, and sectoral laws for AI-relevant provisions. Where regulatory capacity is limited, strengthening enforcement of existing frameworks may be more effective than creating new institutions.
The importance of sector-specific approaches. Policymakers can develop tailored guidance for priority areas (e.g. agriculture, health, public administration, and financial services) and equip sectoral regulators with targeted capacity-building support to improve both relevance and uptake.
The centrality of data quality and governance. Foundational data infrastructure is increasingly recognised as a public investment priority rather than a purely technical concern. Policymakers can explore national open data policies, interoperability standards for government datasets, and the curation of African-language corpora as shared digital public goods that generate returns across multiple sectors.
Regulatory experimentation and risk-based approaches. Adopt risk-based regulatory frameworks that calibrate oversight requirements to the severity of potential harms.
Regional co-ordination as a response to scale constraints. Given the scale constraints facing individual countries, regional co-ordination e.g. through the AU and regional economic communities offers a practical pathway, particularly on shared compute infrastructure, common benchmarks and standards, and cross-border data-sharing frameworks. Governments should designate focal points for regional AI co-ordination and actively participate in these processes.
AI literacy as a component of governance capacity. Mainstream AI literacy into civil service training programmes and national education curricula. Invest in public AI-awareness campaigns. Develop dedicated AI competency frameworks for regulators and policymakers in priority sectors.
The value of structured peer exchange. Institutionalise regular structured peer-learning mechanisms among African AI policymakers, at both bilateral and multilateral levels, to share implementation experience, co-develop solutions to shared challenges, and reduce the duplication of effort across countries at similar stages of AI governance development.
African Union (2024), Continental Artificial Intelligence Strategy, https://au.int/sites/default/files/documents/44004-doc-EN-_Continental_AI_Strategy_July_2024.pdf.
African Union (2013), Agenda 2063: The Africa We Want, https://au.int/en/agenda2063/overview
Department of Communications and Digital Technologies - Republic of South Africa (2024), South Africa National Artificial Intelligence Policy Framework, https://www.dcdt.gov.za/sa-national-ai-policy-framework/file/338-sa-national-ai-policy-framework.html
Federal Ministry of Communication, Innovation and Digital Economy of Nigeria (2025), National Artificial Intelligence Strategy (NAIS), https://ncair.nitda.gov.ng/wp-content/uploads/2025/09/National-Artificial-Intelligence-Strategy-19092025.pdf.
Lesotho (2025), Artificial Intelligence Policy and Implementation Plan, https://aipolicyportal.org/states/lesotho
Masakhane (2026), African Languages Shaping the Future of AI, https://www.masakhane.io/masakhane-african-languages-hub
Ministry of Communication, Telecommunications and Digital Economy of Senegal (2023), Stratégie Nationale et Feuille de Route du Sénégal sur l’Intelligence Artificielle, https://www.archives.sn/docs/strategies/srategie-nationale-intelligence-artificielle
Ministry of Communications and Digitalisation with Smart Africa, GIZ FAIR Forward, and The Future Society (TFS) (2023), Republic of Ghana National Artificial Intelligence Strategy: 2023-2033, https://drive.google.com/file/d/1BBOCB6r6qERMt0lzpzGC-fl2yS0aaMTd/view
Ministry of Communications and Digitalisation with Smart Africa, GIZ FAIR Forward, and The Future Society (TFS) (2023), Republic of Ghana National Artificial Intelligence Strategy: 2023-2033, https://drive.google.com/file/d/1BBOCB6r6qERMt0lzpzGC-fl2yS0aaMTd/view
Ministry of Digital Transition and Digitalisation of Ivory Coast (2024), Stratégie nationale de l’Intelligence Artificielle, https://www.telecom.gouv.ci/new/uploads/publications/174196670372.pdf
Ministry of ICT and Innovation of Rwanda (2022), The National AI Policy, https://www.minict.gov.rw/ai-policy
Ministry of ICT, Postal and Courier Services of Zimbabwe (2026), Zimbabwe National Artificial Intelligence (AI) Strategy, https://veritaszim.net/sites/veritas_d/files/Zimbabwe%20National%20Artificial%20Intelligence%20Strategy.pdf
Ministry of Technology and Science of Zambia (2024), National Artificial Intelligence Strategy (2024–2026), https://www.mots.gov.zm/wp-content/uploads/2025/02/Zambia-Ai-Strategy-Book-option-2.pdf
Ministry of Digital Transformation, Innovation and Modernization of Administration (2024), The National Artificial Intelligence Strategy of Mauritania, https://mtnima.gov.mr/sites/default/files/Mauritani%20AI%20Strategy%20Draft%20April-2024.pdf
Ministry of Digital and Digitalisation of Benin (2023), National Artificial Intelligence and Big Data Strategy 2023 – 2027, https://numerique.gouv.bj/assets/documents/strategie-nationale-d'intelligence-artificielle-et-des-megadonnees-2023-2027.pdf
Republic of Kenya (2025), Kenya Artificial Intelligence Strategy 2025–2030, https://ict.go.ke/sites/default/files/2025-03/Kenya%20AI%20Strategy%202025%20-%202030.pdf
The National Council for Artificial Intelligence (2025), Egypt National Artificial Intelligence Strategy Second Edition (2025-2030), https://ai.gov.eg/SynchedFiles//en/Resources/AIstrategy%20English%2016-1-2025-1.pdf
The National Council for Artificial Intelligence (2020), National Artificial Intelligence Strategy, https://ai.gov.eg/Egypt%20National%20AI%20Strategy-%20English.pdf
The National Agency for the Promotion of Scientific Research (2022), National AI Strategy: Unlocking Tunisia’s capabilities potential, https://www.jaist.ac.jp/~bao/AI/OtherAIstrategies/National%20AI%20Strategy:%20Unlocking%20Tunisia%E2%80%99s%20capabilities%20potential.%20%E2%80%93%2 0Agence%20Nationale%20de%20la%20Promotion%20de%20la%20Recherche%20scientifique.pdf
Working Group on Artificial Intelligence (2018), Mauritius Artificial Intelligence Strategy, https://ncb.govmu.org/ncb/strategicplans/MauritiusAIStrategy2018.pdf
Chaar, T. et al. (2025), “AI and the global productivity divide: Fuel for the fast or a lift for the laggards?”, OECD Artificial Intelligence Papers, No. 51, OECD Publishing, Paris, https://doi.org/10.1787/c315ea90-en.
OECD/LEGAL/0449, Recommendation of the Council on Artificial Intelligence, https://legalinstruments.oecd.org/en/instruments/oecd-legal-0449.
OECD (2025), Scoping note for an AI Policy Toolkit to Support Economies in Realising AI Benefits, https://www.oecd.org/content/dam/oecd/en/events/2025/06/mcm/MCM-2025-Scoping-Note-for-an-AI-Policy-Toolkit-to-Support-Economies-in-Realising-AI-Benefits.pdf
OECD (2024), OECD Artificial Intelligence Review of Egypt, https://www.oecd.org/en/publications/oecd-artificial-intelligence-review-of-egypt_2a282726-en.html.
Phillips, J., A. Safwat and J. Berryhill (2025), The OECD-African Union AI Dialogue 2.0: From strategy to implementation, https://oecd.ai/en/wonk/the-oecd-african-union-ai-dialogue-2-0-from-strategy-to-implementation.
This work is issued under the responsibility of the Secretary-General of the OECD, and does not necessarily reflect the official views of OECD Member countries.
This case study was supported by financial and technical contributions from the Agence Française de Développement and the Foreign, Commonwealth & Development Office of the United Kingdom, in particular for the organisation of the workshop in Nairobi (Kenya) held in March 2026.
This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area.
Photo credits: © Andrey_Popov/Shutterstock.
© OECD 2026
Attribution 4.0 International (CC BY 4.0)
This work is made available under the Creative Commons Attribution 4.0 International licence. By using this work, you accept to be bound by the terms of this licence (https://creativecommons.org/licenses/by/4.0/).
Attribution – you must cite the work.
Translations – you must cite the original work, identify changes to the original and add the following text: In the event of any discrepancy between the original work and the translation, only the text of original work should be considered valid.
Adaptations – you must cite the original work and add the following text: This is an adaptation of an original work by the OECD. The opinions expressed and arguments employed in this adaptation should not be reported as representing the official views of the OECD or of its Member countries.
Third-party material – the licence does not apply to third-party material in the work. If using such material, you are responsible for obtaining permission from the third party and for any claims of infringement.
You must not use the OECD logo, visual identity or cover image without express permission or suggest the OECD endorses your use of the work.
Any dispute arising under this licence shall be settled by arbitration in accordance with the Permanent Court of Arbitration (PCA) Arbitration Rules 2012. The seat of arbitration shall be Paris (France). The number of arbitrators shall be one.