Country Profile on AI Governance: South Africa

Mean of all 10 indicator scores. Southern Africa mean 2.04 · all 54 states mean 1.76.

1. S1.1

   National AI strategy existence

   Score 2 / 4

   National AI Policy Framework published Aug 2024. A full Draft National AI Policy was gazetted 10 Apr 2026 for public comment but was WITHDRAWN on 26 Apr 2026 after it was found to contain fabricated, AI-hallucinated references (two officials suspended). In May 2026, a 7-member expert panel chaired by Prof Benjamin Rosman (Wits) was appointed to redraft the report due Aug 2026, revised draft to Cabinet Nov 2026, public comment targeted early 2027. As of Jun 2026 only the 2024 Framework stands; no binding national AI policy is adopted.

   Note: Framework adopted (2024); full policy absent, draft withdrawn Apr 2026; expert panel redrafting (Cabinet target Nov 2026).

   Sources gov.za (opens in a new tab)

2. S1.2

   Strategy adoption date

   Score 0 / 4

   Policy Framework formally published Aug 2024. Full National AI Policy: not yet adopted (draft 09 Apr 2026).

   Note: Only the 2024 Framework is adopted; full National AI Policy not adopted (draft withdrawn Apr 2026).

   Scoring basis Threshold applied. Raw value: none (adoption date). No formally adopted national AI strategy. Raw value read from the evidence: "Full National AI Policy: not yet adopted (draft 09 Apr 2026)." [notes say the Aug 2024 Policy Framework is "adopted"; could be read as 2024-08]. Checked against the evidence on 7 October 2026: confirmed.

   Sources ellipsis.co.za (opens in a new tab)

3. S1.3

   Political commitment signals

   Score 3 / 4

   Strong senior political commitment: President Ramaphosa announced and established the G20 Task Force on AI during SA's 2025 G20 Presidency; Minister of Communications & Digital Technologies Solly Malatsi opened the G20 AI Task Force (Cape Town, Sep 2025) and the 2024 National AI Government Summit.

   Note: Formal ministerial statements plus presidential/institutional action via the G20.

   Sources weforum.org (opens in a new tab)

4. S1.4

   Dedicated AI budget allocation

   Score 2 / 4

   National AI governance budget identified. AI work funded within DCDT/ICT and university AI hubs.

   Note: Embedded in ICT budget.

   Sources oecd.ai (opens in a new tab)

5. S1.5

   AU Continental AI Strategy alignment

   Score 2 / 4

   SA led the AU AI blueprint (2021) and contributed to the AU Continental AI Strategy (2024), which references continental alignment.

   Note: Partial alignment, no completed formal alignment process yet.

   Sources au.int (opens in a new tab)

6. S1.6

   Multi-stakeholder consultation

   Score 3 / 4

   AI Summit (Apr 2024) engaged industry, academia and civil society, 32 submissions on the Framework, Draft Policy open for public comment.

   Note: Formal consultation with ongoing comment mechanism.

   Sources derebus.org.za (opens in a new tab)

7. S1.7

   Sector prioritisation

   Score 2 / 4

   The 2024 National AI Policy Framework names priority sectors for AI public-value creation, healthcare, education, financial services, agriculture and public services, but without detailed per-sector roadmaps or targets.

   Note: Specific sectors named; detailed sector roadmaps not yet published.

   Sources techpolicy.press (opens in a new tab)

8. S1.8

   Implementation roadmap

   Score 2 / 4

   Draft policy describes a phased adoption approach with implementation projected 2027/2028, Framework itself is high-level.

   Note: Phased plan, accountability/milestones not yet detailed.

   Sources fasken.com (opens in a new tab)

9. S1.9

   Monitoring and evaluation framework

   Score 1 / 4

   The systematic M&E framework is not operational.

   Note: No operational M&E framework; M&E mechanisms only proposed in policy documents.

   Sources dcdt.gov.za (opens in a new tab)

10. S1.10

    Strategy review mechanism

    Score 1 / 4

    The draft policy uses deliberately open-ended language to allow iterative engagement, but there is no formalised, scheduled review mechanism yet.

    Note: Ad hoc / adaptive intent.

    Sources ellipsis.co.za (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 1.43 · all 54 states mean 1.32.

1. G2.1

   AI-specific legislation

   Score 2 / 4

   Draft National AI Policy (Apr 2026) is a precursor to possible future AI law/Act. Existing laws (POPIA, ECTA, Cybercrimes Act) apply to AI indirectly.

   Note: AI law under discussion via the policy process.

   Sources gov.za (opens in a new tab)

2. G2.2

   Data protection law status

   Score 4 / 4

   Protection of Personal Information Act (POPIA, 2013) fully in force since 1 Jul 2021; actively enforced by the Information Regulator.

   Note: Enforced with operational DPA.

   Sources popia.co.za (opens in a new tab)

3. G2.3

   DPA AI mandate

   Score 2 / 4

   Information Regulator has no dedicated AI mandate, but POPIA s71 governs automated decision-making and the Regulator engages on data-driven processing.

   Note: DPA exists; AI engagement informal / via automated-decision provisions.

   Sources popia.co.za (opens in a new tab)

4. G2.4

   DPA staffing and budget

   Score 3 / 4

   Information Regulator funded via the Justice (DOJ&CD) vote. 92 filled positions, headcount, and ~$7.5 million in the 2025/2026 financial year.

   Note: Operational, state-funded DPA with disclosed staffing/budget; ~92 filled posts and ~USD 7.5m (2025/26).

   Scoring basis Threshold applied. Raw value: 92 (FTE staff).

   Sources itweb.co.za (opens in a new tab)

5. G2.5

   AI regulatory body existence

   Score 2 / 4

   Formal process toward a dedicated AI governance architecture is underway. The bodies proposed in the withdrawn draft (AI Commission, AI Ethics Board, AI Regulatory Authority, AI Ombudsperson, National AI Safety Institute) lapsed with that draft, but in May 2026 Minister Malatsi appointed a 7-member expert panel (chaired by Prof Benjamin Rosman, Wits MIND Institute) to design and redraft the national AI policy and its institutional model, with a revised draft due to Cabinet by Nov 2026. No standing AI regulator is operational yet.

   Note: Institution under formation; Rosman expert panel appointed May 2026, revised draft to Cabinet Nov 2026.

   Sources timeslive.co.za (opens in a new tab)

6. G2.6

   Institutional design model

   Score 2 / 4

   De facto model is DPA-led (Information Regulator) plus sector regulators (ICASA, SARB, FSCA). Draft policy signals a move toward a hybrid/cross-sectoral model.

   Note: Hybrid co-regulatory model emerging but nascent.

   Sources gov.za (opens in a new tab)

7. G2.7

   Malabo Convention ratification

   Score 0 / 4

   South Africa has neither signed nor ratified the AU Malabo Convention, citing compatibility with existing national law (POPIA).

   Note: Neither signed nor ratified the AU Malabo Convention.

   Sources diplomacy.edu (opens in a new tab)

8. G2.8

   Enforcement actions taken

   Score 1 / 4

   Robust POPIA enforcement (enforcement notices, first administrative fine of R5m vs DOJ&CD, WhatsApp settlement) but no AI-specific enforcement action, courts have flagged AI-hallucinated citations.

   Note: Data-protection enforcement strong, AI-specific actions only emerging.

   Sources seesa.co.za (opens in a new tab)

9. G2.9

   Extraterritorial provisions

   Score 2 / 4

   POPIA applies extraterritorially to processing using means located in SA, covering some foreign-operated systems, not AI-specific.

   Note: Partial coverage via data-protection law.

   Sources insideprivacy.com (opens in a new tab)

10. G2.10

    IP framework for AI

    Score 1 / 4

    SA Copyright Act already recognises computer-generated works, AI-specific IP review tied to the long-delayed Copyright Amendment Bill, no AI-specific amendments enacted.

    Note: Review initiated, pre-existing computer-generated-work provision.

    Sources recordinglaw.com (opens in a new tab)

Mean of the 9 scored indicators; 1 carries no score. Southern Africa mean 1.77 · all 54 states mean 1.51.

No score: 1 context only.

1. I3.1

   Data center count

   Score 3 / 4

   South Africa is the largest data-centre market in Africa, hosting more than half of the continent's colocation capacity (Teraco, Africa Data Centres, Vantage, NTT, Open Access, Equinix).

   Note: Largest data-centre market in Africa (>50% of continental colocation capacity).

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources arizton.com (opens in a new tab)

2. I3.2

   Data center capacity (MW)

   Score 3 / 4

   Operational capacity in the hundreds of MW and rising rapidly; Teraco alone has 200MW+ in the pipeline; concentrated in Johannesburg and Cape Town.

   Note: Hundreds of MW operational and rising; concentrated in Johannesburg and Cape Town.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources arizton.com (opens in a new tab)

3. I3.3

   Submarine cable landing points

   Score 4 / 4

   Approximately 10 international submarine cables (e.g., Equiano, 2Africa, WACS, SAT-3, SEACOM) land in SA; landing sites incl. Cape Town, Melkbosstrand, Yzerfontein, Mtunzini.

   Note: ~10 international submarine cables land in SA (Equiano, 2Africa, WACS, SAT-3, SEACOM).

   Scoring basis Threshold applied. 11 cable systems (TeleGeography, accessed 2026-10-06).

   Sources arizton.com (opens in a new tab)

4. I3.4

   International bandwidth (Gbps)

   Score 3 / 4

   Very high international capacity; Equiano (144 Tbps) and 2Africa (180 Tbps design) both serve SA, alongside legacy cables.

   Note: Very high international capacity (Equiano, 2Africa) though exact aggregate Gbps not published.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources arizton.com (opens in a new tab)

5. I3.5

   Hyperscaler presence

   Score 4 / 4

   AWS (Cape Town region, 2020), Microsoft Azure (Johannesburg + Cape Town regions, 2019), Google Cloud (Johannesburg, 2024), plus Oracle and Huawei Cloud.

   Note: 4+ hyperscalers with local regions (AWS, Azure, Google Cloud), plus Oracle and Huawei.

   Sources datacenters.com (opens in a new tab)

6. I3.6

   GPU/HPC access

   Score 3 / 4

   Centre for High Performance Computing (CHPC, CSIR, Cape Town) operates the national Lengau supercomputer; commercial GPU also via hyperscaler regions.

   Note: National Lengau supercomputer (CHPC/CSIR) plus commercial GPU via hyperscaler regions.

   Sources arizton.com (opens in a new tab)

7. I3.7

   Data localisation provisions

   Score 2 / 4

   National Data and Cloud Policy (2024) addresses data classification/residency; POPIA governs transfers; comprehensive AI-specific localisation not in force.

   Note: General framework.

   Sources oecd.ai (opens in a new tab)

8. I3.8

   Cross-border data flow framework

   Score 3 / 4

   POPIA s72 provides a comprehensive legal framework for cross-border transfers of personal information.

   Note: Comprehensive cross-border transfer framework via POPIA s72.

   Sources resourcehub.bakermckenzie.com (opens in a new tab)

9. I3.9

   Foreign AI surveillance spending

   ○Context only: Context flag outside the score

   Documented deployment of AI surveillance. Vumacam LPR/CCTV analytics and SAPS facial-recognition/predictive policing, but procurement spending on these systems is not publicly disclosed.

   Note: Partial coverage, use documented, spending unknown.

   Scoring basis Context flag; outside the score.

   Sources technologyreview.com (opens in a new tab)

10. I3.10

    Sovereign cloud initiatives

    Score 2 / 4

    National Data and Cloud Policy and SITA state data-centre consolidation underpin sovereign-cloud ambitions; a dedicated sovereign cloud is under development.

    Note: Under development.

    Sources polity.org.za (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 1.40 · all 54 states mean 1.20.

1. H4.1

   AI researchers per million

   Score 4 / 4

   SA has the strongest AI research concentration in sub-Saharan Africa (CAIR network, multiple universities) but a per-million researcher figure is not reliably published.

   Note: Strongest AI-research concentration in sub-Saharan Africa (CAIR network); exact per-million figure not published.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources arxiv.org (opens in a new tab)

2. H4.2

   DPA staff with AI competency

   Score 1 / 4

   The number of Information Regulator staff with AI technical skills is not publicly reported.

   Note: No AI-specific technical headcount reported within the DPA.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources itweb.co.za (opens in a new tab)

3. H4.3

   University AI programs

   Score 3 / 4

   Numerous universities offer AI/ML programmes; the CAIR consortium spans 7+ universities; dedicated degrees incl. Sol Plaatje (BSc AI), Wits, Stellenbosch, UP, UCT, UKZN.

   Note: 10+ institutions

   Scoring basis Threshold applied. Raw value: 6 (universities).

   Sources up.ac.za (opens in a new tab)

4. H4.4

   African-authored AI publications

   Score 4 / 4

   SA leads sub-Saharan Africa in AI publication output, strong groups at UP DSFSI, Wits, UCT, Stellenbosch.

   Note: SA leads sub-Saharan Africa in AI publication output (UP DSFSI, Wits, UCT, Stellenbosch).

   Scoring basis Threshold applied. 1027.0 papers/yr ÷ 64.01m people = 16.05 per million (floors applied).

   Sources arxiv.org (opens in a new tab)

5. H4.5

   AI patents filed

   Score 1 / 4

   AI-specific patent filings are low (in the single digits to low tens annually); the domestic AI patenting culture is limited.

   Note: AI-specific patenting is minimal; no reliable published count.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources xlaw.co.za (opens in a new tab)

6. H4.6

   K-12 CS education

   Score 3 / 4

   Coding & Robotics curriculum piloted (2023) and being phased in across Grades R-9

   Note: Phased rollout

   Sources education.gov.za (opens in a new tab)

7. H4.7

   AI workforce training programs

   Score 3 / 4

   Government-supported digital/AI skilling via NEMISA, DCDT initiatives and university programmes; mostly pilot scale.

   Note: Pilot stage.

   Sources dcdt.gov.za (opens in a new tab)

8. H4.8

   Labour transition policy

   Score 1 / 4

   Presidential Commission on the 4IR (2020) flagged labour-market impacts; no dedicated AI labour-transition policy adopted.

   Note: Under discussion.

   Sources gov.za (opens in a new tab)

9. H4.9

   Informal economy AI assessment

   Score 2 / 4

   The published government assessment of AI's impact on the informal economy; draft policy notes uneven distribution of benefits.

   Note: Planned / mentioned only.

   Sources gov.za (opens in a new tab)

10. H4.10

    African language NLP coverage

    Score 3 / 4

    Most of SA's 12 official languages have NLP resources via SADiLaR, Masakhane and Lelapa AI (Vulavula, InkubaLM) - isiZulu, isiXhosa, Afrikaans, Sesotho, Setswana and more.

    Note: Most of SA's official languages have NLP resources via SADiLaR, Masakhane and Lelapa AI.

    Scoring basis Threshold applied. Raw value: 5 of 10 (languages supported / major languages), normalised 50. 50% of major domestic languages. Raw value read from the evidence: "isiZulu, isiXhosa, Afrikaans, Sesotho, Setswana and more" [five named; text says "most" of the official languages (lower bound); denominator (my own knowledge. Checked against the evidence on 7 October 2026: confirmed.

    Sources arxiv.org (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 1.16 · all 54 states mean 1.05.

1. N5.1

   AI companies count

   Score 3 / 4

   SA hosts the largest AI/tech-startup base on the continent (several hundred firms with AI components, e.g. Lelapa AI, Aerobotics, DataProphet)

   Note: Largest AI-company base in Africa; exact count not published.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources saaiassociation.co.za (opens in a new tab)

2. N5.2

   Private AI investment (USD)

   Score 2 / 4

   AI investment is a growing subset of SA's broader startup VC.

   Note: Growing subset of SA's broader startup VC; no consolidated annual AI-investment figure published.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources council.science (opens in a new tab)

3. N5.3

   AI incubators/accelerators

   Score 2 / 4

   Multiple incubators/accelerators with AI focus: AlphaCode, Grindstone, Startupbootcamp AfriTech, university AI hubs, hyperscaler programmes. UNESCO ranked South Africa among the top 20% of countries globally in digital infrastructure

   Note: Multiple active AI-focused incubators/accelerators (AlphaCode, Grindstone, Startupbootcamp AfriTech, university hubs).

   Scoring basis Threshold applied. Raw value: 3 (incubators/accelerators). Raw value read from the evidence: "Multiple incubators/accelerators with AI focus: AlphaCode, Grindstone, Startupbootcamp AfriTech" [three named; lower bound]. Checked against the evidence on 7 October 2026: confirmed.

   Sources tia.org.za (opens in a new tab)

4. N5.4

   Regulatory sandbox (AI)

   Score 2 / 4

   IFWG Regulatory Sandbox (SARB, FSCA, National Treasury) is active, with participants (fintech, incl. AI use cases); no dedicated AI-specific sandbox yet.

   Note: Active fintech sandbox; not AI-specific.

   Sources ifwg.co.za (opens in a new tab)

5. N5.5

   Government AI procurement

   Score 1 / 4

   Government procures AI (SAPS, Home Affairs) on an ad hoc basis; no formal framework or domestic-developer preference policy.

   Note: Ad hoc government AI procurement; no formal framework or domestic-developer preference.

   Sources techandjustice.bsg.ox.ac.uk (opens in a new tab)

6. N5.6

   AI safety evaluation capacity

   Score 2 / 4

   AI safety/evaluation capacity exists in academia (CAIR, CSIR); no accredited national AI safety/testing lab.

   Note: Academic research only.

   Sources cair.org.za (opens in a new tab)

7. N5.7

   ISO/IEC 42001 adoption

   Score 1 / 4

   Awareness of ISO/IEC 42001 (AI management systems) is growing; early/pilot adoption by some firms, but not widespread.

   Note: Aware / early.

   Sources sabs.co.za (opens in a new tab)

8. N5.8

   Standards body AI competency

   Score 3 / 4

   SABS operates a national mirror committee to ISO/IEC JTC1 SC42 (AI); SA actively participates in international AI standards work.

   Note: Active AI technical committee.

   Sources sabs.co.za (opens in a new tab)

9. N5.9

   AI risk assessment frameworks

   Score 1 / 4

   Sector-specific voluntary risk guidance emerging (e.g., financial sector); no cross-sector mandatory AI risk-assessment framework.

   Note: Voluntary guidelines.

   Sources cms.law (opens in a new tab)

10. N5.10

    University-industry AI partnerships

    Score 3 / 4

    Formal university-industry AI collaboration via CAIR, CSIR partnerships and joint programmes with NVIDIA/Microsoft/Google and corporates.

    Note: Joint labs / programmes.

    Sources cair.org.za (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 1.48 · all 54 states mean 1.26.

1. E6.1

   Algorithmic impact assessment

   Score 1 / 4

   The algorithmic impact assessments; draft policy references responsible-AI principles only.

   Note: Voluntary guidance emerging.

   Sources gov.za (opens in a new tab)

2. E6.2

   Bias audit requirements

   Score 2 / 4

   The FSCA/PA 2025 financial-sector report and the 2024 Framework set out expectations to mitigate bias, but as voluntary/principle-based guidance rather than enforceable audit obligations.

   Note: Voluntary/principle-based bias mitigation only; no mandatory audits.

   Sources resbank.co.za (opens in a new tab)

3. E6.3

   Transparency obligations

   Score 1 / 4

   POPIA s71 gives data subjects limited rights regarding automated decision-making; no AI-specific transparency mandate.

   Note: Limited via data-protection law.

   Sources recordinglaw.com (opens in a new tab)

4. E6.4

   Citizen redress mechanism

   Score 2 / 4

   Data subjects can challenge automated decisions under POPIA s71 and lodge complaints with the Information Regulator.

   Note: DPA complaint channel.

   Sources recordinglaw.com (opens in a new tab)

5. E6.5

   Judicial AI remedy

   Score 2 / 4

   Ordinary courts have begun handling AI-related matters (ChatGPT-hallucinated citations, deepfake disputes); no specialist tribunal.

   Note: Emerging cases.

   Sources techandjustice.bsg.ox.ac.uk (opens in a new tab)

6. E6.6

   AI advisory body diversity

   Score 1 / 4

   The PC4IR (2020) had some diversity; no standing, demographically representative national AI advisory body currently operates.

   Note: Body existed / homogeneous-to-some diversity.

   Sources gov.za (opens in a new tab)

7. E6.7

   Global AI governance participation

   Score 3 / 4

   SA is active internationally: G20 Presidency 2025 featured AI/digital; endorsed UNESCO Recommendation on the Ethics of AI; engages UN processes.

   Note: Leadership role (G20 2025 host).

   Sources jgu.s3.ap-south-1.amazonaws.com (opens in a new tab)

8. E6.8

   Civil society AI engagement

   Score 2 / 4

   Civil society engaged via the AI Summit and public-comment processes; engagement formal but not yet co-governance.

   Note: Formal consultation.

   Sources africanphilanthropy.issuelab.org (opens in a new tab)

9. E6.9

   African ethical framework reference

   Score 2 / 4

   The Framework/draft policy references African values and Ubuntu as guiding ethical principles.

   Note: Substantive integration of African ethics.

   Sources journals.sagepub.com (opens in a new tab)

10. E6.10

    Youth participation

    Score 2 / 4

    Youth engaged through summits and communities (Deep Learning Indaba), and a youth participation mechanism in AI governance.

    Note: Youth engaged via summits/communities (Deep Learning Indaba); no formal youth governance mechanism.

    Sources activatechangedrivers.co.za (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 1.62 · all 54 states mean 1.55.

1. R7.1

   AU AI Strategy alignment

   Score 3 / 4

   SA contributed to the AU Continental AI Strategy and is aligning its national policy with it; the alignment process is underway.

   Note: Contributed to the AU Continental AI Strategy; national alignment process underway.

   Sources au.int (opens in a new tab)

2. R7.2

   Kigali Declaration endorsement

   Score 3 / 4

   South Africa is among the 49+ African states (plus the AU and Smart Africa) that endorsed the Africa Declaration on Artificial Intelligence at the inaugural Global AI Summit on Africa, Kigali, 4 Apr 2025.

   Note: Endorsed (Kigali, Apr 2025).

   Scoring basis Official signatory list applied. Signatory of the Africa Declaration on AI (Kigali, 4 April 2025), C4IR list: "Endorsed" = 3.

   Sources c4ir.rw (opens in a new tab) · saaiassociation.co.za (opens in a new tab)

3. R7.3

   REC AI governance coordination

   Score 1 / 4

   SA is a SADC member and the region's largest economy, but there is no formal SADC-level AI governance coordination mechanism; SA's regional AI leadership runs mainly through AU/continental channels.

   Note: SADC member; REC-level AI coordination still nascent.

   Sources carnegieendowment.org (opens in a new tab)

4. R7.4

   Cross-border AI agreements

   Score 1 / 4

   No AI-specific bilateral or multilateral governance agreements identified; cooperation occurs through broader continental instruments (AU Continental AI Strategy, Africa Declaration) rather than dedicated AI treaties.

   Note: No AI-specific agreements; cooperation via continental frameworks.

   Sources michalsons.com (opens in a new tab)

5. R7.5

   Election integrity AI provisions

   Score 2 / 4

   No AI-in-elections legislation, but the IEC issued a Framework of Cooperation with Google, Meta and TikTok plus Media Monitoring Africa to counter (incl. AI-generated) disinformation for the 2024 elections, extended toward the 2026 local elections.

   Note: Institutional guidelines/agreements; no legislation enacted.

   Sources elections.org.za (opens in a new tab)

6. R7.6

   Disinformation response framework

   Score 2 / 4

   Institutional (not legislative) framework: Media Monitoring Africa's Real411 platform plus the IEC's multi-stakeholder Framework of Cooperation and the Information Regulator manage AI-generated disinformation.

   Note: Institutional framework (Real411 + IEC); no standalone AI-disinformation law.

   Sources real411.org (opens in a new tab)

7. R7.7

   AI election incidents documented

   Score 3 / 4

   Documented AI-related election incidents in 2024, e.g. a deepfake video falsely showing Donald Trump endorsing a SA party, and AI-manipulated political clips reported to Real411.

   Note: At least 2 documented AI-disinformation election incidents (2024).

   Scoring basis Threshold applied. Raw value: 3 (rule band). Scored on documented institutional response, not incident count. Raw value read from the evidence: "the IEC issued a Framework of Cooperation with Google, Meta and TikTok plus Media Monitoring Africa" [incidents: at least 2 (2024); covers AI-generated disinfo. Checked against the evidence on 7 October 2026: confirmed.

   Sources eisa.org (opens in a new tab)

8. R7.8

   Platform content moderation

   Score 2 / 4

   Formal engagement with platforms via the IEC Framework of Cooperation (Google, Meta, TikTok; X is not a party), channelling reported content to the IEC's Directorate for Electoral Offences.

   Note: Formal platform requirements via IEC framework; not full enforcement powers.

   Sources elections.org.za (opens in a new tab)

9. R7.9

   Geneva Dialogue participation

   Score 2 / 4

   SA is an active participant in UN AI governance processes (Global Digital Compact; G20 AI Task Force, which welcomed UNGA resolutions A/RES/78/265 and A/RES/78/311). Specific participation in the Jul 2026 UN Global Dialogue on AI Governance is not yet separately confirmed.

   Note: Active in UN AI processes; Geneva Dialogue delegation likely but not yet confirmed.

   Sources g20.utoronto.ca (opens in a new tab)

10. R7.10

    G20/multilateral AI engagement

    Score 4 / 4

    Agenda-setter: as 2025 G20 President, SA established and chaired the G20 Task Force on AI, Data Governance & Innovation, launched the AI Initiative for Africa with the AU and UNESCO, and hosted the first-ever G20 Leaders' Summit on African soil (Johannesburg, Nov 2025), whose Declaration commended SA's AI leadership.

    Note: Agenda-setter, G20 2025 host and AI Task Force originator.

    Sources g20.utoronto.ca (opens in a new tab)

Mean of all 10 indicator scores. Southern Africa mean 0.70 · all 54 states mean 0.78.

1. M8.1

   AI governance budget disbursed

   Score 2 / 4

   National AI-governance budget line; AI work is funded within the DCDT/ICT vote and via the AI Institute of South Africa and university AI hubs. Disbursement against a dedicated AI-governance commitment is not separately reported.

   Note: Embedded in ICT budget; no dedicated AI-governance disbursement reported.

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources oecd.ai (opens in a new tab)

2. M8.2

   Health AI regulation

   Score 2 / 4

   The 2024 Framework names healthcare as a priority sector and academic/clinical literature (SAMJ 2025) calls for ethical frameworks, The national health-AI guidelines or legislation are under discussion.

   Note: Under discussion; no enacted health-AI guidance.

   Sources scielo.org.za (opens in a new tab)

3. M8.3

   Health AI devices approved

   Score 1 / 4

   AI health tools are in clinical use (e.g. CAD4TB TB-screening), but SAHPRA does not maintain a separate AI-device approval series, so no verified count is publicly available

   Note: AI health tools deployed

   Scoring basis Fellow score kept; no raw value for the threshold.

   Sources sahpra.org.za (opens in a new tab)

4. M8.4

   Government AI system registry

   Score 0 / 4

   No public registry of AI systems used by the government has been published.

5. M8.5

   AI governance M&E reports

   Score 0 / 4

   No published M&E reports on AI-governance implementation, consistent with the absence of an adopted national AI policy.

   Note: None published.

   Scoring basis Threshold applied. Raw value: 0 (reports).

6. M8.6

   Agriculture AI governance

   Score 1 / 4

   No sector-specific AI-agriculture governance instrument. The 2024 Framework names agriculture as a priority sector, and agritech (e.g. Aerobotics) is active, but no guidelines or legislation are in force.

   Note: Under discussion; no enacted agriculture-AI guidance.

   Sources techpolicy.press (opens in a new tab)

7. M8.7

   Financial services AI governance

   Score 2 / 4

   Strongest P8 area: the SARB Prudential Authority and FSCA published a joint report on AI in the SA financial sector (24 Nov 2025), signalling principle- and risk-based sector guidance and coordination with the Information Regulator under POPIA; the IFWG sandbox is active for fintech (incl. AI).

   Note: Central-bank/conduct-authority guidance issued; sector-specific AI regulation forthcoming.

   Sources resbank.co.za (opens in a new tab)

8. M8.8

   Education AI governance

   Score 1 / 4

   Institutional guidelines are emerging at the university level (NWU AI Hub, UWC, UJ; coordinated via Universities South Africa), and DHET has a higher-education digital-transformation strategy; school-level (CAPS) guidance continues to grow.

   Note: Institutional guidelines emerging; no national education-AI policy.

   Sources usaf.ac.za (opens in a new tab)

9. M8.9

   Measurable governance outcomes

   Score 1 / 4

   AI-specific protective governance outcomes are anecdotal. Robust POPIA enforcement (fines, settlements) addresses data-driven processing, but no measurable AI-governance protective outcomes have been documented, given that the policy framework has not yet been adopted.

   Note: Anecdotal; AI-specific protective outcomes not yet documented.

   Sources techandjustice.bsg.ox.ac.uk (opens in a new tab)

10. M8.10

    WHO AI health guidance adoption

    Score 1 / 4

    South Africa has not formally legislated a direct "adoption" of the World Health Organisation's (WHO) AI guidelines as a standalone law. However, the WHO's frameworks heavily influence South Africa's emerging health technology regulations, which are currently being localised by the country's major health authorities

    Note: Not adopted; awareness assumed from international engagement.

    Sources bowmanslaw.com (opens in a new tab)