The African Union Adopted An AI Strategy And Then Forgot To Implement It
Fifty-five states endorsed a continental roadmap in 2024, yet roughly half the continent still has no national framework to show for it.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Endorsing a strategy is inexpensive. It requires a session, a communiqué and a photograph. Implementing one requires drafting capacity, institutional design, budget lines and officials who can hold a technical brief. Africa has done the first and, in most cases, not yet the second. Two years after the continent committed itself to a common approach to artificial intelligence, the gap between what was agreed and what exists is wide enough that business should stop planning around the agreement.
CONTEXT AND BACKGROUND
The African Union endorsed its Continental Artificial Intelligence Strategy in 2024, urging member states to craft comprehensive national strategies and governance frameworks and to elevate the technology to a national priority.
The response has been thin. Since 2018, thirteen national AI strategies have been adopted across the region, and those countries that have adopted them face common challenges in operationalising the documents and translating policy objectives into effective governance and implementation. That assessment comes from a policy dialogue convened in Nairobi with senior officials from twelve African governments. Comparative research puts roughly half of African nations still at the early stages of formulating a roadmap at all. Progress among the thirteen is uneven, with some governments considerably further advanced than others, so this is not a picture of uniform inaction.
INSIGHT AND ANALYSIS
South Africa illustrates the difficulty precisely, and it is the continent’s most instructive case because it was not for want of trying. The country produced a full draft national policy running to eighty-six pages, proposing a tiered approach to risk, obligations for developers of high-risk systems, new oversight bodies and a funding mechanism.
The draft was then withdrawn. The Communications Minister confirmed that the document published for public comment contained fictitious sources in its reference list, describing the failure as one that compromised the integrity and credibility of the policy. Formal retraction followed by gazette, two officials were placed on precautionary suspension, and revised documents are now expected to reach Cabinet late this year for publication in early 2027.
Reasonable people will read that episode differently. The reading I would offer is that it demonstrates capacity strain rather than indifference. A department attempting an ambitious policy without sufficient specialist drafting capability produced a document its own review layers could not adequately check. That is the operationalisation problem the OECD identified, arriving in visible form.
IMPLICATIONS
Executives should stop treating the continental strategy as a forward indicator. It describes an intention, not an operating environment. For the next several years, most African markets will regulate AI through existing instruments, principally data protection law, rather than through dedicated frameworks.
Boards should therefore assume that the binding constraints on their AI deployments will come from three places: their own governance codes, the data protection statutes already in force, and the extraterritorial reach of European rules that apply to anyone exporting into that market. Waiting for national policy is not a strategy, because in most of the continent there is nothing to wait for.
The most practical lever is procurement. Very few African enterprises train their own models. They buy them, embed them in workflows and inherit whatever the supplier built. That makes vendor due diligence and procurement standards the point at which a board can actually exercise control, and it does not require any national framework to begin.
There is also a harder question underneath. Roughly ninety per cent of the world’s supercomputing capacity sits in advanced economies. Strategies that promise sovereign capability without addressing that concentration are describing an aspiration, and the countries that produced the strongest documents are not necessarily the ones best placed to act on them.
CLOSING TAKEAWAY
The continental strategy was a genuine achievement of coordination, and the criticism here is not of its content. The difficulty is that a framework agreed by fifty-five states and implemented by thirteen is not yet a framework. It is a shared statement of intent, and businesses that mistake one for the other will misjudge the environment they operate in.
The useful response is neither to dismiss the continental project nor to overstate it. It is to recognise that capacity, not ambition, is the binding constraint, and that building the ability to draft, review and enforce technical policy is slower and less visible work than announcing a strategy. It is also the only work that will make the announcement mean anything.
Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments