From Soweto to the frontier: AI close to home

Two weeks after the Trump administration ordered Anthropic to take its most powerful AI models offline, the US company's negotiations with Washington remain unresolved, according to The Verge. Anthropic has sent senior staff to Washington but has declined to give any public update on the talks, and it is unclear when or whether the models will return. South African businesses, researchers, and developers who rely on Anthropic's tools face potential disruption, and the episode raises a broader question about how much control governments can exercise over access to frontier AI systems.

Khoi Tech, a startup founded in Soweto in 2020, has shifted from making consumer electronics such as wireless earphones and smartwatches to developing AI-powered health and workplace technology, TechCabal reports. The company now builds platforms for remote patient monitoring, employee wellness, and driver fatigue management, using artificial intelligence (software that learns patterns from data) to turn health information into early warnings and insights for employers, healthcare providers, and logistics operators. Founder Seati Moloi represented the company at London Tech Week for a second consecutive year as one of nine South African startups selected by the UK-SA Tech Hub, and says the company is in discussions with venture capital investors and exploring expansion across Africa and into the United Kingdom.

ITWeb reports that South Africa's financial services sector is turning to AI-based tools to counter AI-enabled fraud and cyber threats, a shift that reflects how the same technology is being used both to attack and to defend financial systems. The development is relevant to South African consumers and institutions, as wider use of automated decision-making in security and fraud detection raises questions about accuracy, accountability, and who bears the cost when these systems get it wrong.

Google's search business is still growing, but AI tools that answer questions directly, rather than returning a list of websites, are beginning to change how people find information online, according to Memeburn. The shift matters for South Africa because Android, which runs on roughly 77% of South African mobile phones according to StatCounter data cited in the article, gives Google a strong foothold in local AI use, but it also means any drop in search-driven traffic hits South African businesses, publishers, and advertisers directly. Separately, the UK's Competition and Markets Authority has ordered Google to make its search rankings fairer and more transparent, and while that ruling does not apply here, Memeburn notes that regulatory changes in major markets often influence how products are designed globally.

Prosus, the Amsterdam-listed technology group with South African roots, plans to bring a product called ToqanClaw to South Africa, according to Business Day. The company describes ToqanClaw as a way to connect AI models to real-world business applications, though the report provides limited detail on how the product works or which sectors it targets. South African businesses considering AI adoption may want to watch how this offering develops and what it costs in practice.

South African tech outlet Memeburn reports that companies worldwide are pulling back on open-ended AI use after discovering that token-based billing (where costs accumulate each time an AI model reads or writes a piece of text) can spiral quickly when large numbers of staff use these tools for routine tasks. Real examples include Uber capping employees at $1,500 a month on AI coding tools and GitHub moving all its Copilot plans to usage-based pricing from June 2026. Memeburn flags a particular risk for South African businesses: most AI tools are priced in US dollars, so rand-denominated budgets take a harder hit when usage goes unmanaged, and PwC Africa data cited in the piece shows 72% of South African companies plan to expand AI use in the next 12 months.

Anthropic, the company behind the Claude AI system, has launched a feature called Claude Tag that lets workplace teams bring Claude into Slack conversations by typing @Claude, allowing the AI to read channel discussions, complete tasks, and retain context over time. The feature is currently in beta for business and enterprise customers, with plans to expand to other platforms. South African companies considering the tool will need to weigh it against POPIA, the country's privacy law, which requires organisations to apply safeguards to personal information and cannot treat workplace data as casual input for an AI system, according to Memeburn.

The Bank for International Settlements, the international body that coordinates central banks worldwide, has warned that rising government debt, uncertainty around artificial intelligence, and persistent inflation together pose a serious threat to global economic stability, according to Business Day. For South Africa, which carries significant sovereign debt and whose financial system is exposed to global capital flows, a destabilising shift in international markets could tighten borrowing conditions and slow growth. The BIS warning adds institutional weight to concerns that AI's uneven and unpredictable economic effects may amplify existing financial vulnerabilities rather than simply deliver productivity gains.

Researchers at North-West University argue in a new paper that AI companions, chatbots, and social robots can ease loneliness and provide support, but fall short of what genuine human relationships offer. Drawing on the French philosopher Paul Ricoeur's account of happiness as something built through real, mutual connection, the authors contend that AI systems cannot act from their own will, hold moral responsibility, or share authentic experiences, and that substituting them for human bonds risks undermining well-being over time. The paper, republished via The Conversation on Stuff South Africa, contributes a local academic voice to a global debate about how AI is reshaping friendship, intimacy, and community.

Elihle Stali, a student from the Eastern Cape, has built a pair of smart glasses that use artificial intelligence (software that processes information to assist with tasks like recognising surroundings) to help visually impaired people navigate independently, according to IOL. The project was inspired by his late grandmother's blindness and has drawn international attention, with a crowdfunding campaign on BackaBuddy exceeding its target. The story illustrates how South African students are applying AI to local, community-rooted problems in assistive technology.

Writing in Business Day, commentator Ghaleb Cachalia argues that traditional media has been hollowed out by cost-cutting and an editorial drift toward clicks over depth, with AI and platform algorithms accelerating the erosion of public trust in news. For South Africa, where a healthy independent press is central to democratic accountability, the piece raises questions about what is lost when the institutions that verify and contextualise information are weakened. The opinion does not present new data, but it reflects a debate that is live in South African media circles.

Business Day columnist Arthur Goldstuck examines how mobile network operators are under growing pressure to adopt AI-driven, self-managing network systems as demand from AI applications strains existing infrastructure. The piece, by one of South Africa's best-known technology analysts, points to the architectural changes telecoms providers here may need to make as AI workloads grow. For South African consumers and businesses, how well local networks adapt will shape the reliability and cost of the connectivity that underpins AI services.

China's LineShine supercomputer has taken the top spot on the June 2026 TOP500 list, the public ranking of the world's fastest computing systems, beating the US machine El Capitan on a standard scientific-computing measure. Memeburn notes, however, that LineShine ranked only fourth on a benchmark designed to reflect the kind of calculations used in training and running AI models, and that the largest AI computing clusters operated by major cloud companies are often not submitted to the list at all. For South Africa, the outlet argues the more pressing question is access to the physical infrastructure AI depends on: data centres, reliable electricity, and affordable computing capacity, areas where local energy constraints already create real limits.

OpenAI has unveiled its first custom processor, called Jalapeño, built with semiconductor company Broadcom and designed to handle AI inference – the computing work that happens each time an AI tool responds to a user. According to Reuters, as reported by Memeburn, OpenAI plans to deploy the chip by the end of 2026 as part of a broader push to control more of its own computing infrastructure, reducing dependence on dominant chip supplier Nvidia. For South Africa, Memeburn notes that as major AI companies lock advanced computing capacity into large private infrastructure deals overseas, local businesses, developers, and universities accessing AI through cloud services may find that pricing and availability are shaped by decisions made far outside the continent.