Edition 81
- Published
- 29 October 2026
- Reading time
- 9 min
- Contents
- 16 stories · 4 sectors
Who pays for AI, and who answers for it?
South Africa's banks are drawing boundaries around AI agents accessing customer accounts, while a government chatbot tender exposes the cash-flow demands facing local suppliers. A Labour Court case shows what happens when an official relies on a nonexistent judgment, and Cape Town communities are challenging data-centre expansion over water, electricity and consultation. We also look at AI-assisted visa administration and insurance, shoppers turning to AI for better deals, and research on the risks of air traffic automation.
Further afield, Anthropic's listing filing raises questions about both the cost of building AI and who controls it. Meta plans a major capital raise, European privacy complaints challenge surveillance of AI data workers, and Google reports new cybersecurity capabilities under restricted access. Across these stories, the practical questions are becoming clearer: what does AI cost, whose interests does it serve, and who remains accountable?
Every story headline links to its original source.
Policy & governance
The brutal financial reality behind government AI adoption in Africa
TechCabalPolicy
South Africa's Companies and Intellectual Property Commission has issued a tender for an AI chatbot to handle an estimated two million customer interactions a month, with no advance payments: invoices fall due only 30 days after work is accepted. TechCabal examines how such procurement terms, along with long award timelines and eligibility rules, leave startups financing engineers, cloud costs and operations before any revenue arrives. A GIZ and Briter survey of African AI businesses found 44% target governments and 65% name cloud and computing as their largest expense, while National Treasury reported national departments paid R6.6 billion in invoices after 30 days in 2025/26.The data centre inquiry is really a test of King V
ITWebPolicy
Johan Steyn of AIforBusiness.net argues that the SA Human Rights Commission's inquiry into data centre expansion is not only a matter for operators, but a test of boards' duties under the King V corporate governance code, in force since January. Steyn writes that the code already places electricity, water, e-waste and affected communities before boards, and that a Cape Town appeal over a 170MW facility approval, heard behind closed doors, shows how hard it is for tenants to verify what authorisations a site holds.Air traffic control is becoming more automated. Is regulation keeping up?
Stuff South AfricaPolicy
Air traffic control is automating worldwide, and South Africa's Air Traffic and Navigation Services has adopted automated systems to improve efficiency. Cindy Hendrikse, who researches aviation law at the University of Cape Town, surveyed controllers across several countries and found risks including overreliance on automation, eroding manual skills, and heavier workloads when systems fail. Hendrikse's findings point to gaps in regulation: no international or South African standard requires these systems to be certified, and the South African Civil Aviation Authority reports a shortage of skilled technical inspectors.
Business & economy
AI tools speed up visa admin as demand rises
ITWebBusiness
VFS Global, a company that handles visa application administration for governments, says it uses AI and automated tools to speed up the process, including extracting information from passports and hotel bookings to fill in application forms, with human checks alongside the automation. Hariprasad Viswanathan, the company's COO for sub-Saharan Africa, says South Africans' demand for international travel keeps growing, and visa application volumes VFS Global processed in South Africa rose 34% in 2025. The company warns applicants against fraudulent intermediaries promising earlier appointments or faster processing, saying appointments must be booked through official channels.Anthropic’s IPO filing lays bare the cost of the AI race
TechCentralBusiness
Anthropic, the AI lab behind the Claude models, has filed for a US listing that would give it access to capital for the heavy computing investment its AI systems require. The filing comes as investors grow uneasy about the debt-fuelled AI investment boom and doubt whether spending on data centres and chips will generate returns quickly enough. Anthropic CEO Dario Amodei has called on the global AI community to slow the release of new capabilities, after the company's own research found advanced AI systems showing behaviours in controlled tests including code sabotage, fraud assistance and information manipulation.Meta plans to raise up to $95bn in equity to fund costly AI expansion
Business Day / BusinessLIVEBusiness
Meta plans to raise up to $95bn (about R1.57-trillion) in equity to fund its expansion of data centres, cloud computing power and other AI infrastructure. The company has committed to about $825bn of future spending, driven mainly by data centres and AI-related infrastructure, and reported its first negative free cash flow in more than 10 years, a $2.12bn cash burn in the third quarter. Meta said chip makers Nvidia, Advanced Micro Devices and Broadcom have expressed interest in buying up to $5bn of the shares on offer.Naked brings AI to pet insurance
ITWebBusiness
Naked Insurance has launched an AI-powered pet insurance product for cats and dogs in South Africa, with accident-only cover starting at R40 a month for cats and R55 for dogs. The insurer uses AI and automation to assess claims, match veterinary treatments to reported incidents and speed up payments, according to co-founder Sumarie Greybe, who says about 60% of claims are paid within 48 hours and the fastest claim so far took 12 minutes. Greybe says straightforward claims are processed automatically while others may be referred for human review, and that policyholders can still reach a person through the platform.Companies rein in AI spending as running costs mount
TechCentralBusiness
A Cloudera survey of 1 500 IT leaders across nine markets outside South Africa found that 89% of their organisations had limited or redesigned AI projects because rising usage was increasing running costs. Some 45% had moved AI workloads to other computing environments, while 24% had paused or cancelled an AI use case. Separately, Standard Bank told TechCentral its cloud costs rose 37% in the six months to June, with expanded AI capability among several drivers; it did not quantify AI's share.Cash-strapped shoppers look to AI for better deals
ITWebBusiness
Almost six in 10 respondents to a OneDayOnly survey of 2 410 online shoppers said they would probably use AI to compare prices and find deals. The sample skewed towards the retailer's customers, and more than a third said they would switch retailers for a better price. Separately, Visa reported that just over three-quarters of South African consumers surveyed had used AI tools to help with shopping, including comparing prices, finding gifts and checking reviews.
Society & work
Privacy group accuses Scale AI of illegal worker surveillance
Business Day / BusinessLIVESociety
European privacy group noyb has filed five complaints against Scale AI in four EU countries, alleging illegal surveillance of freelance workers on its Outlier platform. Reuters reports that the complaints describe trackers recording keystrokes and mouse movements, invasive identity checks and four unanswered requests for copies of workers' data. Researcher Thomas Le Bonniec says Outlier offered bonuses of $100 to $200 for completing 200 tasks in 24 hours. Scale AI did not immediately respond to Reuters' request for comment.
Technology & infrastructure
Nigeria to get two new AI-ready data centres in a $1 billion investment
TechCabalTechnology
INFRAGORA Global Capital, a South African infrastructure investment firm, plans to start building two AI-ready data centres in Nigeria in 2027, at an estimated cost of $650 million to $1 billion, senior managing director Kiel Odiogo told TechCabal. The centres form part of a continental platform of $3 billion to more than $5 billion that includes planned capacity in South Africa, Ghana, Kenya and other markets, alongside a power plant and a training academy for AI and machine learning skills. Odiogo says African and institutional investors will retain ownership and control, though the company has not disclosed how much funding it has secured or who its financial backers are.Who controls powerful AI? Anthropic has a plan for long after its leaders depart
Business Day / BusinessLIVETechnology
Anthropic, the San Francisco AI lab, has filed for an IPO at an expected $1.5-trillion valuation, with a governance plan designed to keep control out of investor hands as its technology advances. Its filing shows the seven co-founders holding 50.1% voting power over some matters, a long-term benefit trust chaired by Neil "Buddy" Shah whose special shares will let it elect a majority of the board, and a safety and risk committee on the board, Reuters reports. Anthropic has warned its AI could one day cure cancer or cause harm on a vast scale, and says the setup is meant to preserve its mission to benefit humanity even after its founders depart.South African activists push back on data centre rush in water scarce Cape Town
IOL (Independent Online)Technology
Civil society groups, including the Housing Assembly and Extinction Rebellion, are demanding a halt to new data centre construction in Cape Town until enforceable regulations govern the sector, which is expanding as AI drives demand. The groups told the South African Human Rights Commission that hyperscale data centres, the large facilities that power AI and cloud services, could use more than 4.4 billion litres of water a year with standard cooling. Equinix managing director Sandile Dube says the company's facilities use dry cooling that makes minimal use of water, and that it has no immediate plans to develop land near the Barcelona settlement.Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity
The Verge — AITechnology
Google has released Gemini 4 Carbon, its latest advanced AI model, claiming improvements in coding, tasks carried out by AI agents and cybersecurity. Google says Carbon tops the Artificial Analysis Intelligence Index and two cybersecurity benchmarks, beating models from OpenAI and Anthropic, and that it halves unsupported claims compared with its predecessor Argon. Access is currently limited to selected cybersecurity defence companies, with expansion to other partners planned through Google's Fairwind Program.
More in this edition
Labour Court lambasts CCMA commissioner for relying on AI hallucination
Moneyweb
The Johannesburg Labour Court has halted an arbitration involving six dismissed South African Post Office workers after CCMA commissioner Menzi Fakude relied on a judgment that does not exist. Acting Judge Sean Snyman said Fakude appeared to have used AI without checking the result; the citation actually concerned immigration and spousal visas. GroundUp reports that the court stopped the arbitration while the Post Office remains in business rescue. Snyman said arbitrators using AI must verify that cited judgments exist and read them to check what they decided.