For years, South Africa has taken a wait-and-see approach to regulating artificial intelligence. There were, arguably, more pressing legislative priorities in the past, with laws aimed at socio-economic development taking precedence. That grace period has run out amid the exponential rise in AI technology.
Sector-specific and intergovernmental efforts on responsible and ethical AI use have emerged, yet none carry the force of law. Some hold the view that AI should not be regulated because it is not fully understood. The longer we wait, the more exploitation and rights violations we will witness.
South Africa is not lacking in laws. It lacks uniformity and coherence. While examining multiple approaches to AI regulation, it became clear that fragmentation, poor implementation and long-standing governance silos enable AI misuse to flourish in South Africa.
This can also be characterised by overlapping jurisdictions, conflicting mandates, confusion over who authorises what and multiple bodies claiming a stake in the same terrain.
Where a gap exists, it will, in all probability, be exploited. This is why the country needs a constitutional approach to AI regulation. A practical, hybrid framework that closes the gaps by establishing genuine interdisciplinary oversight, clarifying conflicting mandates and avoiding the duplication of governance structures.
Every legal framework is shaped by the Constitution, which exists to guide what a functional, rights-respecting society should look like.
The mandate does not pause for new technology. As former Justice Sisi Khampepe observed in a 2017 Constitutional Court judgment, the ambit of the rights the Constitution protects must take account of developments in technology, particularly where the state has endorsed the technology, just as changes in our social structure reshape the ambit of our rights.
AI regulation should be approached through the lens of what any constitution ultimately seeks to protect. This has practical implications for how oversight is built.
South Africa has constitutionally established institutions and sector-specific regulators, including our chapter nine institutions.
The question is whether they can be properly capacitated to oversee AI in their sectors. For instance, if the South African Health Products Regulatory Authority has oversight over human and animal health, can a similar risk-classification capacity be built for AI?
The point is not to reinvent the wheel but sustainably adapt to what exists.
AI principles and self-regulation have their place and have helped fill some gaps. But industries tend to adopt the principles that are easiest to fold into their mission or vision, while setting aside harder commitments. Self-regulation alone cannot adequately address exploitation where the law is silent. The state has a duty to enact enforceable laws that respect and uphold human rights to close the loopholes.
The risks posed by AI misuse are not hypothetical. AI-enhanced cybercrime is driving data breaches and theft in the country.
Systems trained on data with little relation to South African or even African realities are being deployed in ways that entrench racial and gender inequality.
Deepfakes threaten privacy, creative work is harvested without consent and surveillance is expanding across neighbourhoods with little accountability.
Since AI systems are trained on the data made available to them and our own data on race, gender, disability and culture has long been skewed, the tools risk reproducing and amplifying biases that predate them. Those most exposed are often the most vulnerable: along lines of race and gender, across the digital divide and among children.
A sustainable framework needs to be formalised as binding law, not left to shift based on the goodwill of sectors/industries. It needs to be purposive, built around what South Africa is trying to protect rather than imported wholesale from elsewhere.
It requires interdisciplinary input at every stage. Our courts understand this logic through assessors, trained experts who sit alongside a judge during certain trials. The roles exist because no judge can reasonably be an expert in everything.
AI regulation demands the same humility, with technical experts able to explain what the law cannot see on its own and help decide when a matter calls for sector discretion versus a clear, universal rule.
Companies developing or deploying AI systems should be legally required to demonstrate compliance with human-rights standards, not as an annual tick-box exercise but as a means of accountability across a system’s full life cycle, from creation to decommissioning.
That might raise objections but it is a proactive safeguard for responsible use.
The fact remains that there is no simple formula for regulating AI without discouraging investments and innovation.
African states face pressure to make concessions in line with an international regulatory landscape shaped elsewhere, though some interoperability might be unavoidable.
However, there is a difference between effective regulation and perfunctory regulation that offers no real protection for rights.
For example, regulatory sandboxing, combined with a human-rights lens, offers one way to strike that balance, encouraging innovation while preserving rights protections. Nor should South Africa simply import a European or American model.
The African Union’s Continental AI Strategy rightly insists on context-specific development tha people-centred, inclusive and col orative. South Africa can draw on best approaches while shaping so thing that reflects its own const tional values. A regulatory appro that is distinct yet connected to global conversation.
For South Africans, who incr ingly interact with AI tools with necessarily knowing it, a formal framework should mean clear r for introducing AI tools for gen use, accountability throughout t life cycle, accessible complia mechanisms and public awarenes all 12 official languages.
Authentic information is pow Amid genuine “AI fatigue” a bombardment of AI slop, it mains essential to know where data is going, how it will be p cessed and what legal recourse ex if mishandled.
The saying “doing what you with what you have” could go a l way towards that, while plans being made to establish a possibl regulator, equip regulators, colla rate with experts across multiple ciplines and explore multiple reg tory approaches remain key.
South Africa has waited l enough. The Constitution provid framework for action. What rema is the will to formalise it, in a m ner that upholds the rule of and human rights while leav room for the innovation the coun badly needs.
Nokuthula Olorunju is a PhD graduate from the University of Cape Town.