Building, using, and being cut off from AI

The South African Revenue Service (SARS) has issued a call for proposals to overhaul its systems using artificial intelligence (AI) – software that can automate analysis and decision-making – within four months, according to Business Day. The agency wants to reduce its dependence on manual processing of tax work. If implemented, the change would affect how millions of South African taxpayers and businesses interact with the country's main tax collector.

The US government issued an export control directive to Anthropic, the company behind the Claude family of AI models, ordering it to suspend access to its newly released Mythos 5 and Fable 5 models for any foreign national, including Anthropic's own foreign-national employees, according to The Verge. Anthropic determined the only way to comply was to shut the products down entirely, and travelled to Washington to contest the order. For South Africa, the development is a signal that access to frontier AI models, the most capable systems available, may increasingly be subject to geopolitical controls that cut off local researchers, developers, and businesses without warning.

An AI-powered debt-collection system operating in South Africa is reportedly recovering R10 million per month, according to ITWeb. The story points to growing use of automated systems in the local credit and financial-services sector, where AI handles tasks traditionally done by human collectors. How such systems treat consumers, and what oversight applies to them, are questions with direct relevance to South Africans carrying debt.

Visa is preparing South African banks to work with AI shopping agents – software that can browse, select, and pay for goods on a customer's behalf – according to TechCentral. The company has also struck a partnership with OpenAI, the organisation behind the ChatGPT service, which TechCentral reports adds further momentum to the rollout. For South African consumers and banks, the development raises practical questions about how automated purchasing decisions will be authorised, disputed, and kept secure.

Google's managing director for Africa, Alex Okosi, told TechCabal that African startups are already building AI into core products, but investment and infrastructure have not kept pace. Speaking at the graduation of 15 startups from Google's African accelerator programme in Nairobi, Okosi pointed to gaps in cloud capacity, data centres, and funding as the main constraints on scaling AI across the continent, noting that Africa holds only about 1% of global data centre capacity. South Africa features directly in the picture: a local transit startup called Loop is among the cohort, Google is investing in a cloud region in the country, and a $5.8 million grant programme targets AI skills development for civil servants and nonprofit leaders in South Africa, Kenya, and Nigeria.

Dr Miné de Klerk, dean of curricula and research at South African private higher education provider Eduvos, argues in Moneyweb that AI is changing what employers expect graduates to know, shifting the focus toward ethics, critical thinking, and responsible use of the technology. The piece raises questions about how universities and colleges should update their programmes to prepare students for workplaces where AI tools are increasingly common. The editor should note the article appears under Eduvos's own section on Moneyweb, which may indicate sponsored or branded content.

South African startup Supascale AI has launched what it describes as Africa's first cloud marketplace for AI computing power, according to IT News Africa. The platform lets businesses, developers, and creators access AI models (the large software systems that power tools like chatbots and image generators) at lower cost, while owners of graphics processing units (GPUs, the specialised hardware used to run AI workloads) can earn money from capacity they are not using. If the cost and access claims hold up, the model could lower one of the main barriers to building and using AI locally, though the startup's own assertions about pricing and market position have not been independently verified here.

SoftBank, in partnership with OpenAI, has launched a cybersecurity service in Japan called Patching as a Service, which uses AI to help organisations find security weaknesses in their systems and plan how to fix them before attackers can exploit them. The service initially targets Japan's largest critical infrastructure operators, including airports, power systems, and banks, with a team of around 50 people set to grow to 1,000. South Africa has its own legal framework around cyber risk, including the Cybercrimes Act and a Reserve Bank directive requiring payment institutions to protect critical systems, so how AI-powered cyber defence tools are governed, priced, and distributed across organisations of different sizes is a question that will eventually land here too.