🏛️ Policy, Institutions, and Talent

📙 The Strategy Landscape, Calibrated

By July 2025, sixteen African countries had published national AI strategies, with several more in draft. The continental layer is anchored by the AU Continental AI Strategy (July 2024) and the AUDA-NEPAD Continental AI Roadmap (February 2025). Reading these as a single landscape rather than as separate national documents, the most useful empirical observation is that the policy-to-legislation pipeline is almost entirely broken. Only Ethiopia has drafted AI legislation; that draft has not been enacted. South Africa's policy was withdrawn before it could take effect (see the next section). Every other major African economy regulates AI use, where it regulates it at all, through adjacent data-protection statutes inherited from the GDPR family.

This is the operational form of the Yilma & Wodajo critique. The argument is not that the strategies are insincere or empty — many are detailed, ambitious, and explicitly grounded in African political-economic positions — but that strategy is not the same thing as governance. Governance requires binding instruments. The strategies, with extremely limited exceptions, do not yet have any.

A consolidating analytical resource: Munga & Quansah's September 2025 Carnegie piece Understanding Africa's AI Governance Landscape compares 15 national + 2 continental strategies and offers the cleanest single map of the policy stack. Engida Abdella & Alayande's April 2025 piece for the Global Center on AI Governance, African Countries Are Racing to Create AI Strategies — But Are They Putting the Cart Before the Horse?, provides the calibrated counterweight: many strategies skip foundational infrastructure questions, lack realistic implementation timelines, and rely on benefit-sharing rhetoric without operational mechanisms. CIPESA's 2025 SIFA report (Navigating the Implications of AI on Digital Democracy in Africa) draws on empirical research across 14 countries and is the strongest single empirical complement to the analytical work.

🇷🇦 The South African AI Policy Withdrawal

The most consequential recent event in African AI governance happened in South Africa in April 2026, and it deserves treatment in detail because it is simultaneously a perfect example of the Yilma & Wodajo critique and a worked instance of the policy-and-practice problem we covered in 11.3. It also — for the students reading this from a UCT vantage point — happened in your own jurisdiction.

There are three reasons this case is worth a careful read.

• It is the exact failure mode Week 11.3 covered, at policy-document scale. The He & Bu finding was that policies don't curb the surge in AI use because researchers don't disclose. Here, an entire government ministry submitted an AI-drafted policy document without disclosure or citation verification, and the failure was caught not by the ministry's internal processes but by a journalist with thirty minutes to spare. The policy that was meant to govern AI use in South Africa was itself an instance of the problem it was meant to address.
• It is the strongest available South African instance of the Yilma & Wodajo critique. The argument is that strategy is not governance. Here, even the strategy document collapsed before it could attempt governance. The country is not currently between strategies; it is in a documented policy vacuum, with the previous draft withdrawn and no successor published.
• It is unusually concrete material for a postgraduate course. The full gazetted PDF is publicly available. The withdrawn citations are documented. The methodology that produced the problem is the methodology this entire course is asking you to be cautious about. If you want a single piece of evidence for why the calibrated-reading habit matters, this case is it.

🏛️ Continental Bodies, and What They've Actually Done

Three pan-African or African-adjacent bodies are worth knowing about, and one obvious confusion to head off at the outset.

💰 Who Is Actually Paying for African AI?

The honest answer in May 2026 is: Big Tech capex and Northern foundations. The sovereign and continental instruments are largely announced; the operational funding flows are largely commercial. This is the central empirical observation underpinning everything in the second half of this sub-lesson.

🏫 The Institutional Landscape

Beneath the policy layer, the institutional infrastructure that trains and retains African AI researchers is genuinely substantial. It is also concentrated, and where it isn't concentrated the gap is large. Six institutions or programmes do the disproportionate share of the work.

A note on national AI institutes. NCAIR (Nigeria, commissioned November 2020) is the most-cited national institute, though as we noted in 11.4 it does not have substantial dedicated AI compute. The Ethiopian AI Institute (EAII, inaugurated November 2025 under Director-General Dr Worku Gachena) is the implementing body for the National AI Policy and is the strongest example of a national institute moving toward operational delivery. C4IR Rwanda is operational and co-designed Rwanda's AI policy; the Rwanda AI Scaling Hub (with Gates Foundation, US$7.5M MoU April 2025) sits there.

🖋️ The Talent & Brain-Drain Question

The talent pipeline is the layer of the African AI picture where the data are weakest and the rhetorical claims are strongest. The honest position is that no rigorous AI-specific quantitative study of the African brain-drain problem exists yet. What exists is general migration data (Africa loses around US$2 billion per year in skilled migration; UK granted 78,000+ work visas to Nigerians by September 2023) which does not disaggregate AI researchers from the broader skilled-worker population. When you see specific figures for “African AI talent loss”, treat them with the same calibration the rest of this week has been asking for.

What we can say more confidently:

• The pipeline is top-heavy. Indaba and AMMI feed a substantial pool of MSc-level talent into the system. The PhD bottleneck is where the system leaks hardest — both because African PhD positions in AI are scarce and because the relative value of doing a PhD at, say, MIT or Mila is large for the individual researcher even when the African opportunity cost is also large.
• Diaspora is the unstated default. The Africa Declaration on AI explicitly bundles “Africa and diaspora” in its proposed AI Scientific Panel composition. The 2025 Qhala / CMU / C4IR AI Research Colloquium drew 100+ researchers “from continent and diaspora”. The proportional split between Africans on the continent and Africans in diaspora is not publicly quantified, but the rhetorical and policy framing treats them as a single research community.
• Some retention levers are working. Sovereign compute access (the UCT ACI, the Rwanda AI Scaling Hub) is explicitly framed as anti-brain-drain. Named chairs (Marivate's ABSA Chair of Data Science at UP), substantial corporate grants (Google PhD Fellowships at Wits, the US$1M to AfriDSAI), and partnerships with industry (Lelapa × Microsoft) all provide structural alternatives to leaving. Salary parity with US or UK industry positions is impossible; competitive intellectual conditions are not.
• Some notable returns happen. Moustapha Cissé returned to Senegal to start KERA Health. Vukosi Marivate did a US PhD at Rutgers and returned to UP. Raesetje Sefala, DAIR's first fellow, is based in Johannesburg. These are not statistically significant; they are existence-proof signals that the flow is not one-way.

🔍 Where the Gaps Are

Five gaps stand out from the policy-and-institutions picture. Some are obvious. Others are easy to miss precisely because they map onto where the existing infrastructure is.

1. Geographic concentration. The substantive PhD quadrilateral is in South Africa. The strongest national-institute work is in Ethiopia, Nigeria, and Rwanda. Beyond this small set, dedicated AI research infrastructure is sparse to absent. Francophone West Africa outside Senegal and Rwanda, Central Africa (DRC, CAR, Chad), the Horn of Africa outside Ethiopia, and Lusophone Africa (Angola, Mozambique) have near-zero structural AI research capacity as of May 2026.
2. The policy-to-legislation gap. Already noted: only Ethiopia has draft AI legislation; only one African economy is actively legislating; the rest is strategy. If you are interested in AI law and policy as a research area, this is a wide-open field.
3. Operational continental funding. The AU AI Grant (US$100M) and Investment (US$200M) mechanisms are uncapitalised. The Africa AI Fund (US$60bn) is a pledge envelope. The IDRC, Lacuna, and Google.org flows are all operationally important but are not sovereign in any of the senses 11.4 set out.
4. Audited KPI tracking. Strategies announce headline targets (Egypt: 30,000 AI professionals by 2030; Rwanda: US$76.5M policy implementation; Nigeria: 70% AI literacy in 16–35-year-olds by 2030). Publicly auditable tracking dashboards against these targets are largely absent. This is true across the continent, not just for African countries; it is also the gap that distinguishes a strategy from a programme.
5. The brain-drain data itself. The single most important talent-related research opportunity for an African postgraduate is an actual, rigorous quantitative study of African AI researcher migration. The work has not been done. The general migration figures circulating are insufficient.

🎯 What This Means for Your Research

• Vet supervisors, not institutions. Outside the substantive PhD quadrilateral, African AI supervision is highly individual-dependent. If you are choosing where to do your PhD, the named supervisor matters more than the institution's overall reputation.
• The compute story is now the talent story. Several higher-education-led compute initiatives on the continent (the UCT African Compute Initiative; the Rwanda AI Scaling Hub) are explicit retention levers for African AI talent: the case for them is not only access to GPUs but the institutional argument that the infrastructure the next decade of African AI work will run on should be built and stewarded on the continent.
• Read the strategies. Read them critically. The Yilma & Wodajo paper is the single best entry point to the analytical literature on the strategy landscape. The Carnegie and Global Center pieces are the complementary practical reads. The SA AI Policy case is the worked example of why this matters at the policy-document level.
• Brain-drain framing is the wrong default. The honest pattern is not “Africa loses its AI talent to the North”; it is “African AI work is distributed across continent and diaspora”. If you are an African postgraduate researcher, your collaborative network is plausibly larger and more international than the brain-drain rhetoric implies. Build it accordingly.
• The work matters. An institutional landscape with this many gaps means that a postgraduate who chooses to work on one of them can plausibly produce a contribution that the field will recognise. Week 12 closes the course by asking what you would actually do with that opportunity.

✏️ A Short Exercise

1. Pick the African country you would most likely engage with in your current or planned research project (this may be South Africa by default, but does not have to be).
2. Find its current AI strategy or policy document. If it has been withdrawn, find that fact. If none exists, note that as your answer.
3. Read the document's preamble and pillars. Identify the strategy's stated goals and the timeline to which they are committed.
4. Identify one specific commitment in the document that a postgraduate researcher could imaginably contribute to. It could be a dataset, a benchmark, a piece of empirical research, an institutional contribution, anything concrete.
5. Bring it to class. Together with the data-and-models exercise from 11.5 and the disclosure-policy exercise from 11.3, you will have the raw material for the Week 12 capstone pitch.

📚 Sources & Further Reading

Coming up in Week 12: the integrative capstone for the course. We synthesise the futures half (11.1, 11.2) and the African-capacity half (11.4–11.6) and ask the question that has been quietly underlying everything in the second half of the course: what is the most useful thing a postgraduate in this room can actually do? The Week-12 page carries the full brief for the 3-hour solo capstone activity.