Governance questions pile up as AI spreads

ITWeb reports that the Information Regulator, the body that oversees access to personal information, has set out the regulatory questions that artificial intelligence raises. The article covers how the regulator is engaging with the governance challenges AI poses for South Africa.

The City of Cape Town has acknowledged its approval process for large data centre applications needs new rules. A data centre is the building and computing equipment that stores data and runs AI systems, and a major application by the company Equinix is currently suspended. The outcome will shape how quickly AI infrastructure can be built in the city.

China has rejected calls by US AI leaders, including Anthropic's Dario Amodei, to slow the development of frontier AI models, with a Foreign Ministry spokesman saying that fearmongering and confrontation would only disrupt global AI governance, Memeburn reports. The piece notes that President Xi Jinping has offered BRICS members, a group that includes South Africa, support to build open-source AI, positioning China as a supplier of the technology rather than a gatekeeper. For South African businesses and government, the dispute matters because it shapes how available and affordable leading AI models will be, with some Chinese models already priced well below their US rivals.

The United Nations human rights chief, Volker Turk, has warned that artificial intelligence could pose an existential threat to humanity, and that people face unfamiliar and unprecedented risks to their rights, IOL reports. The warning comes from a global governance conversation that South Africa takes part in through the UN, and it points to questions about how AI systems should be governed to protect people's rights here.

Artificial intelligence can now play a role in triggering a competition-law dawn raid, an unannounced inspection by competition authorities. The article says South African companies face growing risks around digital evidence, meaning records held in their systems could come under scrutiny in such investigations.

South African company boards are growing more dependent on AI systems without weighing the costs, the risk of being locked into one supplier, or the fact that foreign providers could switch the technology off. The TechCentral piece argues this leaves local companies exposed on decisions that should be made at board level.

South Africa's reading challenges have sharpened debate about how artificial intelligence could support learning in schools. The discussion speaks to a live question for the country's education sector: whether AI tools can help learners who struggle with reading, and under what conditions.

The South African government has mapped 95GW of offshore wind potential, but no law yet allows anyone to lease the seabed to build it. The story links this to the energy demands of AI data centres, the large computing facilities that run AI systems, which need reliable power at scale. The gap between the mapped potential and the missing legal framework is a constraint on where AI infrastructure could be built in South Africa.

Microsoft has released a draft code of conduct that would place fixed safety limits on its future MAI AI models, including rules the models must never help build weapons, must never resist shutdown, and must stay under human control; the draft is open for public comment for six weeks. According to Memeburn, the move matters for South African companies because AI agents (systems that can act across tools and services rather than only produce text) with access to email, customer data or payment systems need clear permissions, audit logs and shutdown controls as part of basic security. The article notes that other leading AI companies, including Anthropic and OpenAI, have separately called for slower capability development until stronger safety standards exist.

Stuff South Africa republishes an analysis from The Conversation on advertising being built into AI chatbots, with OpenAI testing ads in ChatGPT and Google testing sponsored answers in Gemini. The authors argue that paid influence could shape the answers AI systems give, and that current laws and fact-checking tools, built for public social media posts, cannot easily catch misleading claims made in private, personalised AI conversations. For South Africans who increasingly use AI chats instead of search, this raises questions about whether sponsored content could steer what they see, and about the rules needed to keep AI answers trustworthy.