Who benefits, who is exposed, and who decides

The South African Revenue Service is expanding its automated tax assessment system, which uses AI to file returns on behalf of qualifying taxpayers without manual input, as part of a broader technology overhaul following a record R2-trillion tax collection year, according to Business Day. The move will affect a growing number of South Africans who currently receive or will receive a pre-populated return from SARS rather than completing one themselves. How the system handles edge cases and whether taxpayers understand their right to correct an automated assessment are questions the expansion is likely to bring into sharper focus.

When the US government ordered AI company Anthropic to cut off foreign access to two of its most capable models, citing national security concerns, Indian businesses and policymakers were left without tools they had built workflows around. Memeburn reports that the episode has sharpened India's debate about sovereign AI, meaning a country's ability to run critical AI systems without depending entirely on foreign companies, clouds, or rules. The article draws a direct line to South Africa, noting that local companies similarly rely on imported chips, foreign cloud platforms, and overseas AI tools, and referencing the withdrawal of South Africa's first draft national AI policy after it was found to contain apparently AI-generated fictitious references.

MTN is deploying technology that uses its underground fibre-optic cables as vibration sensors, detecting illegal digging before a cable is cut, according to TechCabal. The system is live in South Africa, alongside software agents that monitor network faults and adjust settings automatically, though a human must approve any changes before they take effect. MTN says network and IT operations account for roughly 55% of its operating costs, so the efficiency gains from these tools are central to its stated target of R30 billion in value from AI over the next three to five years.

Oracle shed roughly 21 000 employees last year, a 13% fall in its total workforce, with the company attributing part of the reduction to automating work with AI, according to TechCentral. The development is relevant for South Africa, where Oracle has a local presence and where the broader question of how large technology companies manage AI-driven job cuts is increasingly part of labour and business conversations.

Microsoft is considering adding DeepSeek, a Chinese-developed AI model, to its Copilot Cowork product as a lower-cost alternative to the OpenAI and Anthropic models it currently uses, according to reporting by Axios cited in Memeburn. Copilot Cowork is a tool built into Microsoft 365 that can carry out multi-step tasks across apps such as Outlook, Word, and Teams with limited human input, and running it at scale is proving expensive. For South African organisations that already rely on Microsoft 365, the shift could make this kind of automated work more affordable, but Memeburn notes that questions about data governance, POPIA compliance, and the geopolitical origins of the DeepSeek model will need answers before businesses deploy it on sensitive documents or client information.

South Africa's gap in who benefits from generative AI (AI systems that produce text, images, or other content) is growing along lines of age and education, according to TechCentral. The outlet reports that without deliberate intervention, the technology is likely to deepen existing advantages for those who are already connected, educated, and comfortable with digital tools. This matters because it points to a structural risk: the people with the least access to AI's potential gains may also be the least positioned to advocate for themselves in decisions about how the technology is rolled out here.

Radio is Africa's most widely used source of political news, and Business Day reports that broadcasters across the continent are poorly equipped to detect or counter coordinated AI-generated disinformation, sometimes called 'AI swarms', where automated systems produce and spread false content at scale. The piece argues that this gap leaves a critical public information channel exposed at a time when such tools are becoming cheaper and easier to deploy. For South Africa, where radio reaches communities that other media do not, the integrity of that channel has direct consequences for informed democratic participation.

A Pew Research Center survey of more than 5,000 US adults, conducted in February 2026, found that only 16% believe artificial intelligence will have a positive effect on society over the next 20 years, even as nearly half now use AI chatbots regularly. Memeburn draws a line to South Africa, noting that the same tools shaping American workplaces and classrooms are already present here, and that local public confidence in AI took a knock earlier this year when the government withdrew a draft national AI policy after it was found to contain what appeared to be fabricated references. The survey's core finding, that widespread use does not automatically produce public trust, is relevant context as South Africa works toward a revised policy draft expected in early 2027.

South African universities are moving away from relying on AI detection tools – software used to identify whether student work was written by an AI system – according to ITWeb. The shift signals a broader rethink of how institutions govern AI use in academic settings, with implications for assessment, academic integrity policy, and students across the country.