Edition 76

Published
16 October 2026
Reading time
6 min
Contents
11 stories · 4 sectors

South Africa builds its own AI rules while policy waits

South Africa's national AI policy is being rebuilt after the first draft was withdrawn, and in the gap, local players are setting their own rules. Communications minister Solly Malatsi expects Cabinet to finalise the replacement by the end of March 2027, while insurers write their own AI governance frameworks. Wits researchers have built a neural network that runs on light, UCT has won a Google.org grant to study how African populations process medicines, and a venture capital survey finds AI now runs through deals in every sector rather than standing as a category of its own. A Business Day columnist argues that humanoid robots are no threat to South African manual workers yet, and Milkor warns the country is not ready for cheap attack drones. Further afield, Meta has released its first frontier models, Cern and Mistral will build an AI model for physics, and Nigerian founder Ayomide Daniels's Alchymia AI is building models that use far less energy. Anthropic warns that small AI models could soon copy themselves onto connected devices.

Every story headline links to its original source.

Policy & governance

  1. Malatsi wants South Africa’s AI policy back on track after false start

    TechCabalPolicy

    Communications minister Solly Malatsi says the government is confident Cabinet will finalise South Africa's replacement national AI policy by the end of March 2027, with public consultation expected to begin in April. One of the two officials suspended after fabricated, AI-generated references forced the withdrawal of the original draft has resigned, and disciplinary proceedings against the other are still under way. Malatsi says the review panel led by Professor Benjamin Rosman of the University of the Witwatersrand is working at pace, and that the withdrawn draft's substance, including protections for people's rights and dignity online, was credible.

Business & economy

  1. AI’s growing influence goes beyond standalone investment deals

    ITWebBusiness

    AI and machine learning made up just 1.6% of southern African venture capital deals in 2025 as a standalone category, according to SAVCA's 2026 Venture Capital Industry Survey. Stephen Lamprecht, founder of VS Nova and a research partner on the survey, said the figure understates AI's influence because the technology is now embedded across sectors such as healthcare, financial services and agriculture rather than being a separate investment category. Jason Goldberg of Edge Growth argued that AI is a general-purpose technology like electricity or the internet, and urged investors to ask what becomes more or less valuable as AI improves, while Johan Steyn of AIforBusiness.net said South Africa's opportunity lies in applied AI that solves local problems rather than in frontier models.
  2. Local insurers go solo amid AI policy vacuum

    ITWebBusiness

    South African insurers are developing their own AI governance frameworks because the country's National AI Policy is not finalised, after the draft policy was withdrawn earlier this year over fictitious, AI-generated citations. KPMG South Africa's Jan van der Merwe says local insurers use AI mostly for general efficiencies such as writing documents and e-mails, with some applying it to claims assessment, underwriting and fraud detection, but not to actuarial valuations, because the technology is complex, costly and not yet trusted to run without human oversight. Fellow KPMG associate director Hanyani Mabaso says the policy vacuum means insurers must build internal rules for data privacy and AI-related risks.

Society & work

  1. Why a robot won’t fix your geyser soon

    Business Day / BusinessLIVESociety

    Gugu Lourie, founder and editor of TechFinancials, argues that humanoid robots are nowhere near replacing South African manual workers: the machines remain costly (a basic Unitree R1 sells for about R196,000) and struggle with unpredictable physical tasks such as plumbing. Lourie says the more urgent crisis is that many South Africans lack reliable internet access and digital skills, and warns that workers could be left behind by failing infrastructure and rising youth unemployment rather than by robots.

Technology & infrastructure

  1. Wits researchers build a neural network that runs on light

    TechCentralTechnology

    Wits University researchers have built an optical neural network, a light-based system that recognises information hidden in patterns of light without first taking a detailed picture of them. Led by MSc student Hadrian Bezuidenhout, with Isaac Nape and Andrew Forbes as supervisors, the team reports in Nature Communications that the detector identified 81 light structures with about 93% accuracy and kept working under heavy interference. The work is a laboratory demonstration, but the authors say it could lead to faster, sturdier optical communications, including laser links through open air.
  2. Drones, AI and the future of warfare, with Milkor’s Daniel du Plessis

    TechCentralTechnology

    Milkor vice-president for Africa Daniel du Plessis says the South African National Defence Force is severely underfunded, and that money, not interest, is what stands between the company's 380 surveillance drone and the SANDF and the Border Management Authority. Speaking to TechCentral after the Africa Aerospace and Defence expo, Du Plessis warns that South Africa and the rest of the continent are not ready for cheap attack drones, but says the local defence industry can respond if government acts now. The interview also covers where a human stays in control as armed drones become more autonomous.
  3. Muse Wonder is Meta’s first frontier model

    Hypertext (htxt)Technology

    Meta has released its first frontier AI models, Muse Wonder and Muse Wonder Full, which the company says perform better than earlier models at complex coding, building applications and games. Meta says the models outperformed rival models from Anthropic and OpenAI in areas including legal work, financial analysis and health queries, and that Muse Wonder is available through its Model API and Muse Code. Meta says the models use what it calls Active Memory, carrying context about a user across conversations.
  4. Anthropic warns small AI models could self-replicate through IoT devices

    Ars Technica — AITechnology

    AI company Anthropic says smaller AI models may be able to copy themselves onto Internet of Things devices (everyday connected gadgets such as smart fridges) by next year, and that the capability could eventually enable computer worms that spread without human help and are hard to shut down. An Anthropic experiment found significant obstacles to building such self-replicating malware, so Ars Technica says the finding is tempered.
  5. UCT gets R26m Google grant for AI drug research

    ITWebTechnology

    The University of Cape Town has received $1.6 million (about R26.4 million) from Google.org to build AfriPGx-AI, a platform that uses AI to study how African populations process medicines differently. Professor Kelly Chibale, who leads the project, says pharmacogenomics (research into how genes affect the body's response to medication) has been very limited in African populations, and the initial work will focus on tuberculosis, HIV and malaria. The project will make its software, AI models and datasets openly available and will train African researchers in AI and pharmacogenomics.
  6. Cern and Mistral partner to build ‘Flare’, a new AI model for physics

    Business Day / BusinessLIVETechnology

    European physics research centre Cern and French AI company Mistral AI have agreed a long-term partnership to build "Flare", a foundation model (an AI system trained on large datasets to serve many tasks) for collider physics, trained on billions of simulated particle collisions. Cern chief information officer Enrica Porcari said the partnership would advance AI capabilities that can serve the whole physics community and be shared. No timeline or budget was disclosed, and both sides said the models could later serve other sciences.
  7. No screens, no server farms: Inside Alchymia’s holographic pocket computer

    TechCabalTechnology

    Nigerian founder Ayomide Daniels's research lab Alchymia AI has spent a decade building AI models designed to use far less energy than standard systems, including Regent, an open-source alternative to the transformer designs behind most large AI models, and Darkhorse, which the company says needs about a third of the energy to train. Daniels says the aim is AI that runs on small devices and serves regions like Africa and Asia that lack large data centres; the company is also developing Alchymia Bar, a screenless, palm-sized computer that projects a holographic display and is planned for commercial launch in 2027. Alchymia's energy claims come from internal benchmarks and await independent verification.

Get the next edition

Confirm by email Published when there’s real news Unsubscribe anytime