Who governs AI, and who is keeping up

Writing in Business Day, Tayla Pinto argues that South African members of parliament cannot fulfil their constitutional duty to oversee the executive if they lack a working understanding of AI infrastructure – the physical and digital systems that underpin how artificial intelligence is built and run. The piece frames AI literacy not as a technical nicety but as a prerequisite for meaningful parliamentary scrutiny of government decisions involving these systems. For South Africa, the argument points to a gap between the pace at which AI is being adopted in public and private institutions and the capacity of elected representatives to hold those institutions to account.

At the 2026 World AI Conference, Chinese President Xi Jinping set out a proposal for reshaping how artificial intelligence is governed globally, framing it around reducing digital inequality and expanding international cooperation. IOL, which filed the piece under its BRICS section, signals editorial judgement that the proposal carries regional weight. As a BRICS member involved in multilateral discussions on AI norms and standards, South Africa could be drawn into the governance frameworks China is seeking to advance, though what that means in practice remains to be seen.

Standard Bank Group expanded its use of artificial intelligence (AI) across its operations while keeping overall banking technology costs nearly flat, with IT spending growing just 2%, according to TechCentral. Cloud computing costs rose sharply, up 37%, though the bank offset this through lower amortisation charges as older assets were written off. For South Africa, the figures suggest that one of the country's largest banks is deepening its AI deployment without a proportional rise in costs, a pattern that will interest observers watching how AI adoption affects bank operations, pricing, and staffing.

Netstar, a South African vehicle-tracking company, is deploying artificial intelligence (software that learns patterns from data to make predictions) to generate insights about vehicle behaviour and maintenance needs, according to Business Day. The company says processing this data locally, rather than sending it to remote servers, reduces costs and improves the speed of those predictions. The move signals growing adoption of AI within South Africa's vehicle-tracking sector, where Netstar is seeking to recover competitive ground.

Checkers Sixty60 has upgraded its AI shopping assistant, Pixie, to accept text, voice, and image inputs, according to Stuff South Africa. The feature, called Pixie Smart Chat, can generate recipes, convert a photographed handwritten recipe into a shopping list, and identify replacement products from a photo. Shoprite is rolling it out first to Xtra Savings Plus subscribers, meaning most Sixty60 users will need to wait before they can access it.

Discovery Group chief information officer Derek Wilcocks tells TechCentral how the company uses AI to tailor its Vitality health and rewards programme to individual members, and shares his view that AI-assisted coding has real limits that developers should understand. Wilcocks also argues that the stronger case for AI investment at Discovery is growing revenue rather than cutting costs. The interview offers a ground-level account of how one of South Africa's largest financial-services groups is putting AI to work in a product used by millions of local consumers.

Writing in Business Day, columnist Rufaro Mafinyani examines how autonomous AI systems – software that acts on its own without step-by-step human instruction – can be attacked, manipulated, or pushed beyond their intended boundaries, and argues that managing these risks has become a board-level responsibility rather than a purely technical one. The piece frames operational redundancy (having backup systems and processes so that a failure in one does not bring everything down) as a core part of how organisations should govern AI. For South African companies adopting AI, the column signals that security and resilience planning needs to sit alongside strategy, not trail behind it.

African cybercriminals are adopting AI tools faster than the law enforcement agencies pursuing them, according to Interpol's African Cyberthreat Assessment Report 2026, as reported by TechCabal. Interpol found that AI was involved in 55% of cybercrime cases across surveyed African countries in 2025, with criminals using generative AI (software that produces text, images, audio, or video on demand) to craft convincing phishing messages, fake identities, and impersonate executives and officials, while deepfake incidents increased sevenfold in the second half of 2024 alone. The report has direct bearing on South Africa, where southern Africa is identified as a hub for organised scam networks and where a rapidly digitising financial system processed over $1.1 trillion in mobile transactions in 2025, creating a large and growing pool of potential victims and stolen data.

Writing in Business Day, organisational psychologist Gleb Tsipursky argues that South African companies should embed AI skills through what he calls 'workflow apprenticeships': manager-led experiments that teach employees to use AI tools within their actual daily tasks, rather than through standalone training courses. The approach is presented as a practical way to close the gap between AI awareness and real workplace use, at a time when South African employers are under pressure to build AI capability across their organisations. Readers should note the author runs a consultancy in this space, and Business Day flags the piece as opinion.

A growing body of research warns that medical students who use AI tools to generate diagnoses before developing their own clinical reasoning may never fully build those foundational skills, a risk researchers call "never-skilling." A University of KwaZulu-Natal study found ChatGPT useful for practising patient history-taking, while academics from the University of Pretoria and the Colleges of Medicine of South Africa have cautioned in a local journal that introducing AI too early as a substitute for independent reasoning could weaken the judgment doctors need in practice. Both groups point toward the same answer: structured use, where students first demonstrate competency without AI before being taught to critically evaluate what it produces.

Anthropic, the company behind the Claude family of AI models, has announced it will embed invisible, machine-readable watermarks in content processed by its models worldwide, according to Ars Technica. The move is driven by the European Union's AI Act, which requires providers to mark AI-generated or AI-altered audio, images, text, and video, but Anthropic is applying the requirement to all new models globally rather than only in the EU. South African developers and businesses that build on or use Claude-based tools will be subject to this change, and the broader shift toward mandatory AI-content labelling signals a direction that local regulators and companies may need to account for.