South Africa should regulate AI through binding rules anchored in the constitution rather than through voluntary ethical principles and self-regulation, according to a University of Cape Town doctoral thesis.

The argument, by attorney and researcher Nokuthula Olorunju, comes while the department of communications & digital technologies rebuilds the national AI policy along deliberately light-touch lines.

Minister Solly Malatsi has staked South Africa’s AI future on staying neutral between Washington and Beijing, and the redraft sets national guidelines while leaving individual sectors to write their own strategies.

Olorunju does not address the department’s work, but her conclusions certainly run against its grain. Her thesis, Artificial Intelligence, Bias and the Rule of Law: A Constitutional Approach to the Regulation of AI in South Africa, argues that existing legislation, non-binding ethical principles and self-regulation are not enough. Principles lack the force of law, she argues, and self-regulation lets organisations pick the commitments easiest to fold into their own missions.

Fragmentation, weak implementation and legal silos are what allow misuse to thrive, producing overlapping jurisdictions and confusion over which body answers for what. “In essence, if a gap can be exploited, it most probably will be,” she said.

Olorunju, a research fellow at Research ICT Africa, will be capped at UCT on Tuesday, 8 September. Her comments come from a graduand profile published by UCT News on 2 September and a statement issued by the university.

Hybrid model

Her proposal is a hybrid model anchored in the constitution: clarify conflicting mandates, establish an interdisciplinary AI oversight body and avoid duplicating governance structures. Developers and deployers would have to show compliance with human rights standards and accept accountability across a system’s whole lifecycle. Compliance, she argues, should amount to more than an annual “tick-box” exercise.

For ordinary users, the university says that would mean clearer rules governing AI tools, assessments of AI systems, accountability measures and a process for lodging complaints about AI misuse. Olorunju also wants public awareness material available in all 12 official languages.

Existing institutions and sector regulators can carry part of the load, provided they are properly equipped and their mandates clarified.

That does not make her a straightforward critic of South Africa’s delay. UCT casts her argument as a call to move beyond a “longstanding ‘wait-and-see’ approach”. In the profile she is more equivocal.

“There are two sides to this approach,” she told UCT News. “On one hand, South Africa has benefited from seeing how other countries have approached AI regulation; on the other hand, waiting too long could lead to ‘copy-and-paste’ adoption of AI laws that have no reference to the South African context.”

Malatsi withdrew the draft AI policy in April after it was found to cite academic sources that do not exist, then appointed an independent panel chaired by Wits AI researcher Benjamin Rosman to produce a lighter-touch replacement, due at cabinet later this financial year. The delay has already cost visibility, with South Africa absent from a World Bank tally of more than 80 countries with published AI strategies.

Olorunju argues that South Africa should draw on international best practice rather than lifting regulatory models from the US or the EU wholesale, and that because there is no perfect way to regulate AI, countries should start with what they can realistically implement. — © 2026 NewsCentral Media