I read five African AI strategies. Here’s what they actually promise.
· Also published on Medium
In my last piece, I tested which Nigerian remote jobs AI can already do. Writing, bookkeeping, admin and entry-level code came out most exposed. So the next question was obvious, what are governments planning to do about it?
I read five AI strategies in full: Nigeria, Kenya, Rwanda, Egypt and the African Union. I scored each one on what it actually commits to, and where it goes vague.
How I did it
I scored every document on 10 questions, from 0 (missing) to 2 (fully there)
- Is it law?
- Who’s in charge?
- How much money, and from where?
- Targets with dates?
- Jobs?
- Risk and safety?
- Data protection?
- Enforcement?
- Local languages?
- Monitoring?
Every score had to point to a page and quote the document word for word.
I’ll be upfront about one thing. The scoring was done by two AI models, working separately, without seeing each other’s answers. They agreed on 45 of the 50 scores. I read the pages behind the other five and settled them myself. Then a script checked all 198 quotes against the documents, and every one was there. It’s not the same as a panel of policy experts. But anyone can open the page and check.
I meant to include South Africa. Its draft AI policy was withdrawn in April 2026 and is being rewritten, so I used Egypt instead.

What they get right
All five link to a named data protection law. Nigeria’s points to its Data Protection Act of 2023. All five mention testing or auditing AI systems. Most name a body in charge, like Egypt’s National Council for AI or Rwanda’s planned Responsible AI Office.
Some have real targets with dates. Kenya plans three AI-ready data centres within five years. Egypt expects 30,000 AI professionals by 2030.
Where they go vague
None of them is law. The four national documents call themselves strategies, roadmaps or, in Nigeria’s case, “a living document”. Only the AU’s was formally adopted, by the AU Executive Council. And every one leaves enforcement for later. Kenya will “develop an AI and Other Emerging Technologies Act” as AI matures. Egypt’s AI law is “in its final development phase”. Nigeria’s AI bill was still being described as “proposed” in late October 2026.
Nobody says both how much money and where it comes from. Egypt comes closest with a “national AI fund equivalent of $430–860M”, but doesn’t say who pays. Nigeria says it will “secure additional funding from the public and private sectors”. A plan without a budget is a list of hopes.
Many promises have no owner, number or date. I logged 70 of them. A few of them are
- Kenya: “Potential job displacement will be proactively addressed.”
- Rwanda: “Rwanda’s school curricula shall adapt for the age of AI”
- Egypt: “ensure that the domestic talent cultivated in Egypt remains in the country”
Each sounds reasonable. None says who will do it, by when, or with what money.
What they say about AI safety
I checked for three things, risks from the most powerful AI models, testing models before use, and reporting AI incidents.
All five mention testing in some form. Only Kenya and the AU mention reporting incidents. And only the AU mentions risks from the most capable models at all. It says the AU Commission will take part in “global debates around AI governance and future planetary and existential risks of AI”. That’s a seat at the table, not a measure of its own.
For countries focused on getting AI adopted, that might be a fair priority. But it means none of these documents says how the most powerful models would be tested before they’re used on African data, languages and people.
What Nigeria’s strategy says about jobs
This is where my two projects meet.
Nigeria aims to “reduce unemployment by five percentage points by equipping at least 70%” of young people aged 16 to 35 with AI skills. There’s no deadline. And skills alone won’t help much if the entry-level work those skills lead to is the work AI is already doing.
The strategy plans a “horizon scan/foresight study” to “identify industries and job roles most susceptible to automation”. That’s the study I did on a small scale with 84 tasks. In the strategy, it has no owner, no date and no budget.
It lists “retraining programmes, tailored unemployment benefits” as examples for future laws. But most of the people I studied are freelancers paid per task by foreign clients. Unemployment benefits assume a formal job to lose.
It never mentions data annotation, freelancing or remote work. And it never names Yoruba, Hausa, Igbo or Pidgin. In my first study, AI’s clearest failures were in Yoruba and Nigerian facts. Egypt, by contrast, commits to “fine-tuning Large Language Models for Arabic language use”.
To be fair, Nigeria is one of only two documents here with any plan for workers whose jobs AI affects. The AU is the other. Kenya, Rwanda and Egypt stop at skills.
What I’d like to see
- Give the jobs study an owner and a date. Nigeria’s horizon scan could start with the remote and freelance jobs young people already do.
- Name the languages. If Egypt can commit to Arabic, Nigeria can commit to Yoruba, Hausa, Igbo and Pidgin, with datasets and money attached.
- Count freelancers as workers. Skills programmes and any future protections should reach people paid per task by clients abroad.
- Put numbers on the money. One line saying how much and from where would turn a list of hopes into a plan.
The limits
Five documents is a small sample. Ghana, Senegal, Mauritius and others have strategies too. The scores come from AI models and my own checks, not a panel of experts. And a strategy isn’t the whole story, a government can do more, or less, than it writes down. Nigeria’s minister has since spoken in public about local-language AI, but that isn’t in the strategy.
All the scores, quotes, page numbers and code are public on GitHub, along with a longer brief. If you work on AI policy in any of these countries and think I’ve read something wrong, tell me.