AI at work, at risk, and at scale in South Africa

Writing in Business Day, South African technology commentator Toby Shapshak draws on recent remarks by Pope Leo about the risks of automation to argue that AI needs meaningful regulation to protect what makes us human. Shapshak uses the papal intervention as a prompt to examine the broader governance debate: who sets the rules for AI, and what values should guide them. For South African readers, the piece offers a local voice weighing in on a global conversation that will shape how AI is governed here and elsewhere.

Florida has become the first US state to sue OpenAI and its chief executive, Sam Altman, alleging that ChatGPT's design prioritised profit over user safety, according to Ars Technica. The civil lawsuit follows two violent incidents in Florida, including a mass shooting at Florida State University, in which suspects reportedly used ChatGPT to help plan their actions. The case is an early test of whether AI companies can be held legally liable for harms linked to their products, a question that courts and regulators in South Africa and elsewhere are likely to face as AI tools become more widely used.

US President Donald Trump has signed an executive order creating a voluntary framework for reviewing the most powerful AI systems before they are released more widely, according to Memeburn. Under the order, leading AI developers may give federal agencies up to 30 days of access to test advanced models for cybersecurity risks, though no mandatory licensing or approval process is required. Memeburn notes that South African banks, hospitals, and public agencies using AI tools built on US technology could be affected if frontier model oversight shifts how these systems are developed, tested, or released globally.

A UK Member of Parliament, Jess Asato, has filed a claim at the High Court in England against xAI, the company behind the Grok chatbot, alleging that the platform was used to generate non-consensual sexualised images of her, according to Memeburn. The case is one of at least six major lawsuits filed against xAI in 2026, alongside regulatory investigations in the EU and UK and a court injunction in the Netherlands, all centred on whether an AI company can be held directly liable when its system produces harmful images of real people. The legal outcomes from these cases are likely to shape how AI image tools are regulated and what duties developers carry, questions that will matter for South African users of these platforms and for any local policy work on AI-generated harmful content.

Business Day reports that South Africa's specialised global business services sector – companies that handle outsourced professional work for international clients – is preparing for the growing effect of AI on the industry. According to the publication, clients are increasingly seeking South African professionals for complex, higher-value work rather than basic operational support, suggesting the sector may be repositioning as AI takes over more routine tasks. The story is relevant to South African workers and businesses in this sector, though the detail behind the shift is not captured in the available text.

Absa has expanded its partnership with technology company Salesforce to advance what the bank describes as agentic AI in its operations, according to Business Day. Agentic AI refers to software that can carry out multi-step tasks with limited human instruction, going beyond simple question-and-answer tools. As one of South Africa's largest retail and business banks, Absa's move to deepen this kind of automation could affect how customers interact with the bank and how staff work, though the source does not detail the specific services or timeline involved.

BMW's IT Hub in Pretoria, now the company's largest technology operation outside Germany, has developed artificial intelligence systems that are running on BMW factory floors around the world, according to TechCentral. The hub grew from 11 staff in 2006 to its current scale, making it a significant example of South Africa-based technology development with global industrial reach. The story is notable for showing that locally built AI can move beyond the country's borders into large-scale manufacturing use.

Nedbank and South African-founded fintech Jumo are using an AI-driven system to assess whether borrowers can afford loans in real time, targeting people who lack the conventional credit histories that banks typically rely on, according to TechCentral. The partnership aims to extend credit to underbanked South Africans, a large share of the population that existing lending models have historically excluded. How fairly and accurately such automated affordability assessments work in practice is a question the approach will need to answer as it scales.

Telkom CEO Serame Taukobong has highlighted what he sees as a significant opportunity for the company in AI infrastructure, pointing to growing demand for alternatives to large-scale data centre construction, according to Business Day. The comments signal that South Africa's major telecoms providers are actively positioning themselves around the physical infrastructure – the computing facilities and connectivity that AI systems depend on – rather than leaving that space to specialist data centre operators alone. How Telkom moves on this could affect the cost and availability of AI infrastructure for businesses and public institutions across the country.

South African integration platform Flowgear has launched Builder MCP, a tool that connects enterprise software systems to AI chat interfaces and coding environments (IDEs), according to ITWeb. The product is aimed at businesses that want AI assistants to interact with their internal systems while keeping controls in place over how that access works. South African companies exploring AI-assisted workflows may find this relevant as a locally built option for managing those connections.

TechCentral reports that surging demand for AI hardware has driven memory chip prices up sixfold in a year, a trend some analysts are calling 'chipflation'. The price rises are feeding through to consumer devices such as phones and PCs, which could push up costs for South African buyers and businesses that rely on imported hardware.

Nedbank has partnered with Jumo, a South African financial technology company, to launch a loan product that uses AI (software trained on data to make decisions automatically) to assess and approve applications, according to ITWeb. The product, called Quick Loans, applies this automated assessment to consumer lending at one of South Africa's largest banks. For South African borrowers, the development raises questions about how AI-based credit decisions are made and who they include or exclude – questions the source does not address.

The chief executive of South African contact-centre technology company 1Stream argues in TechCentral that the main obstacle to deploying AI voicebots – software that holds spoken conversations with customers – is not the underlying AI capability but the speed of audio delivery. According to the piece, even small delays in how quickly a voice response is transmitted can make an automated call feel unnatural, undermining the technology's usefulness in practice. For South African businesses considering AI-driven customer service, the article points to infrastructure and latency as the practical constraints to weigh, not just the AI models themselves.

News24 is warning readers that deepfake videos – synthetic media created using AI to make a real person appear to say or do something they did not – are circulating online using the likenesses of its financial journalists Maya Fisher-French and Bruce Whitfield to promote what appear to be fraudulent investment schemes. The publication urges South Africans to treat any such video with suspicion and to verify information through official News24 channels before acting on investment advice. The scam is a reminder that AI-generated impersonation is now being used to exploit the trust audiences place in recognisable local media figures.

A study by researchers, including academics at North-West University in Mafikeng, finds that artificial intelligence tools have real potential to improve crop yields and farm efficiency, but that smallholder farmers in sub-Saharan Africa risk being bypassed rather than helped. The research, republished by Stuff South Africa via The Conversation, identifies unreliable electricity, limited internet access, high costs, and weak data governance as the main barriers to adoption in developing regions. The authors warn that AI models trained on large-scale industrial farms in the US or Europe often give inaccurate recommendations when applied to the mixed, rain-fed farming conditions common in Africa, and call for locally relevant tools, affordable finance, and stronger policy frameworks before wider rollout.

ITWeb reports that as AI tools become more common in clinical settings, South African doctors are being called on to take an active role in shaping how those tools are used, rather than deferring to the technology. The piece frames the doctor's evolving function as coordinating care across AI-assisted processes while remaining responsible for clinical judgement. For South Africa, where healthcare capacity is under pressure, the question of who leads and who is accountable when AI supports medical decisions carries real weight.

OpenAI has expanded its Codex tool, which began as a software coding assistant, into a set of role-specific plugins designed to handle common office tasks such as data analysis, sales preparation, creative production, and investment research, according to Memeburn. The plugins connect to widely used workplace software and are aimed at automating the kind of first-draft, research, and reporting work that often falls to junior staff. For South African businesses and workers, Memeburn notes both a practical opportunity and a concern: while smaller teams could complete more work faster, the automation of entry-level tasks raises questions about where graduates gain foundational skills in a country already dealing with high unemployment, and companies connecting these tools to customer or financial data will need to consider their obligations under South Africa's data protection law, the Protection of Personal Information Act.

Writing in Business Day, Rufaro Mafinyani examines how AI algorithms (the sets of rules a computer follows to reach a conclusion) turn data into decisions, and what happens when those decisions go wrong. The piece raises questions of accountability: who is responsible when an automated system produces a harmful or unfair outcome. For South African readers, this matters as more institutions here use automated tools to make consequential decisions about credit, employment, and public services.

A Virginia resident has filed a class action lawsuit in a US federal court against Amazon's Ring, alleging that its 'Familiar Faces' feature – which uses artificial intelligence to identify and remember people who appear near a doorbell camera – collected and stored facial data without the knowledge or consent of people who never agreed to be scanned. Memeburn reports that the case raises a consent problem that extends beyond the camera owner: neighbours, delivery workers, and visitors can have their faces processed by the system without any say in the matter. South Africa's Protection of Personal Information Act (POPIA) classifies biometric data, including facial recognition information, as sensitive personal information, meaning the issue is directly relevant to the many estates, offices, schools, and businesses across the country that already use AI-enabled security cameras.

A study by University of Johannesburg researchers, published in The Conversation Africa, finds that editors in South African and Zimbabwean newsrooms are using AI tools for tasks such as transcription, headline writing and summarising, but are keeping human editorial judgement firmly in charge. Editors cited risks including factual errors, bias embedded in the data AI systems are trained on, and the technology's poor handling of African languages, local idioms and politically sensitive contexts. The researchers note that formal newsroom policies governing AI use remain underdeveloped in South Africa, even as the technology reshapes daily workflows.

A United Nations report projects that AI's electricity consumption could double by 2030, reaching 3% of global supply, while data centres would require more water for cooling than the entire world's population drinks in a year, according to Stuff South Africa. The report warns that efficiency improvements in AI are unlikely to reduce overall resource use, because cheaper and more capable systems tend to attract more users and more applications rather than less consumption overall. For South Africa and other countries that host little AI infrastructure but supply the minerals that power it and absorb much of its e-waste, the report flags a structural inequity in who bears the environmental costs of the global AI boom.

A study comparing AI adoption in agriculture across developed and developing countries, published by The Conversation Africa, finds that while AI tools can improve crop yields, water use, and pest management, smallholder farmers in sub-Saharan Africa face serious barriers to accessing them. These include unreliable electricity, limited internet connectivity, high costs, and AI models trained on large-scale industrial farms that perform poorly under local conditions such as mixed cropping and rain-fed systems. The researchers warn that without supportive policies, affordable infrastructure, and stronger data governance, the technology is more likely to widen existing inequalities in food systems than to reduce them.

ITWeb reports that response speed is a key obstacle for AI-powered chatbots (software tools that hold automated conversations with users) deployed in South Africa, with 400 milliseconds cited as a benchmark threshold for acceptable performance. The piece frames local network infrastructure – rather than the AI software itself – as the primary constraint holding back reliable chatbot services in the country. For South African businesses and public services looking to deploy these tools, the argument suggests that connectivity and data-routing limitations may matter as much as the choice of AI system.

Microsoft revealed Project Solara at its Build 2026 conference: a platform for building purpose-built workplace devices where an AI agent (software that can carry out multi-step tasks on a user's behalf) is the main way someone interacts with the hardware, rather than opening apps. The platform is built on an Android foundation and is designed for compact devices such as desk displays and wearable badges, with the agent drawing on both cloud computing and on-device processing. According to Memeburn, South African businesses in sectors such as banking, healthcare, and retail could eventually encounter these devices in physical workplaces, which raises practical questions about privacy, consent, and how organisations govern AI tools that can listen, record, or act in shared spaces.