The real-world cases behind the modules, each written to match what its source says, with a link to that source. Use these to revisit a case after a session or share with a colleague.
Module 1
Government · South Africa · April 2026
SA Government AI Policy Hallucination Scandal
On 10 April 2026 South Africa’s Department of Communications and Digital Technologies published its draft National AI Policy for public comment. Journalists checked the references and found fabricated ones; the process halted 16 days later when the Minister withdrew the draft, calling it a failure of oversight: generative AI had been used without proper human verification of the sources.
Read: The Conversation →
Module 2
Finance · SA / Nigeria / Kenya · 2025
Credit Scoring: The 37% Gender Penalty
A 2025 audit of 10 credit-scoring algorithms across South Africa, Nigeria, and Kenya found that women-led SMEs received lower funding recommendations, with the study reporting a 37% underfunding penalty — linked to gender signals in ownership, sector coding and networking patterns. No individual loan officer decided to discriminate. The algorithm did it invisibly. Source note: the study used synthetic audit profiles to test algorithmic outputs, not real lending decisions from named lenders. Findings reflect potential algorithmic bias, not a universal estimate for all African fintech lenders. (Dzreke & Dzreke, 2025)
Read: Advanced Research Journal →
Module 2
Finance · Kenya · CGAP, 2016
Kenya Mobile Lending: When Access Becomes the Score
Most Kenyan digital lenders set a credit score and loan size from the borrower’s phone-based data — call and SMS records, mobile money history and social media data — and many operate outside regulation, CGAP reported. Our analysis for the course: a model fed this kind of data can read a thin digital footprint, often a matter of cost or access, as risk, without the person being any less creditworthy.
Read: CGAP →
Module 3
Law Enforcement · South Africa
Predictive Policing: Who Gets Flagged?
A 2024 review of algorithmic bias in Africa reports that predictive policing algorithms in South Africa have been found to target low-income communities, increasing surveillance and harassment of innocent people (citing Singh, 2022). Module 3 explores the mechanism through a labelled composite scenario: when patrols follow past records, the places already policed most produce the most new data.
Read: PMC / Frontiers →
Module 3
Biometrics · Multi-country · 2018
Facial Recognition: Accuracy Is Not Equal
The Gender Shades study tested three commercial gender-classification systems on 1,270 faces of people from Rwanda, Senegal, South Africa, Iceland, Finland and Sweden. Darker-skinned women were the most misclassified group, with error rates of up to 34.7%; the highest error rate for lighter-skinned men was 0.8%. A system like this has to be tested on the people it will actually be used on.
Read: Gender Shades (Buolamwini & Gebru) →
Module 4
Media & Elections · South Africa · May 2024
SA 2024 Election Deepfake Videos
During South Africa’s 2024 elections, deepfake videos were circulated to manipulate perceptions and endorse political entities, CIPESA reports. Post-election research found conventional misinformation was still more widespread than synthetic media — but one convincing fake can spread before anyone checks its source. South Africa has no single “deepfake law”: consequences depend on what the content does and which law applies.
Read: CIPESA State of Internet Freedom in Africa →
Module 5
HR & Legal · South Africa · September 2025
AI Recruitment and the Employment Equity Act
In a September 2025 alert, SA law firm Cliffe Dekker Hofmeyr noted that the Employment Equity Act prohibits unfair discrimination in employment, and that if an AI tool used in recruitment results in discriminatory outcomes, “employers may be held liable.” Their advice: build human review into key stages, especially where AI flags or rejects candidates, and set clear policies on AI use. Not legal advice — take specific questions to a qualified employment lawyer.
Read: Cliffe Dekker Hofmeyr Employment Law Alert →
Module 6
Healthcare · Global, incl. southern Africa · 2021–2025
AI-Assisted TB Screening: A Score, Not a Diagnosis
WHO has recommended computer-aided detection (CAD) software for reading chest X-rays in TB screening since 2021, and in June 2025 reported six products meeting its performance standards for people aged 15 and older. It does not yet recommend CAD for children under 15, and anyone who screens positive still needs confirmatory testing before treatment. Module 6 uses CAD4TB to show AI supporting, not replacing, clinical judgment.
Read: World Health Organization →
Module 6
Education · South Africa · CSIR
Dr Math: Technology That Kept the Human at the Centre
The CSIR Meraka Institute’s Dr Math let learners send maths questions over MXit, a popular chat service, to registered human tutors — mostly University of Pretoria students — with a system routing each request to an available tutor. The emphasis was on imparting skills rather than supplying answers. It won a Technology in Government in Africa award in 2011. Module 6 contrasts it with AI that scores: here the human was the service.
Read: CSIR →
Supplemental: African AI Success Stories
Supplemental
Finance · Africa · Platform for banks and mobile operators
Jumo: AI Behind Mass-Market Credit
Jumo provides a mobile financial technology platform that mobile network operators, payments providers and banks use to offer credit and other financial products at scale. It describes its “leading AI prediction capabilities” as central, says the platform has been behind 250M+ loans, and that its team works across 15 countries on 3 continents.
Visit: jumo.world →
Supplemental
Education · South Africa
Siyavula: Adaptive Practice, Free to Learners
Siyavula offers curriculum-aligned Mathematics (Grades 8–12) and Physical Sciences (Grades 10–12) practice with full solutions and instant feedback, free for learners and data-free on the MTN and Cell C networks in South Africa. Its practice questions are “guided by the algorithm” and adapt to each learner, so each one progresses at their own pace.
Visit: siyavula.com →