South Africa's AI Rules Fracture Into Patchwork; Public Faces Uneven Protections

South Africa’s citizens will navigate a patchwork of AI rules rather than a single national standard, as the government moves away from prescriptive industry-wide regulation toward a sector-led model operating within national guidelines.

The shift carries real consequences for everyday life. AI systems already influence hiring decisions, lending approvals and public administration across the country, yet South Africa currently has no formal national framework governing the technology. The revised approach, presented at the inaugural Naspers/Prosus AI Governance and Policy Symposium in Sandton on Wednesday, asks individual sectors to develop their own AI strategies rather than imposing uniform rules from above.

Jeanette Morwane, acting deputy director-general at the department of communications and digital technologies, explained the reasoning directly. “The whole idea is for the policy to provide guidance and then to get the sectors to start developing their own sectoral AI strategies. We don’t want to prescribe to sectors because they understand the architecture of their sectors. We just want to provide national policy guidelines that they can use,” she said.

The departure from the original draft is substantial. That earlier document proposed creating seven new institutions to oversee AI. Parliament pushed back, and the government now plans to work through existing regulators instead.

The original draft was also withdrawn in April after at least six of the 67 entries in its reference list were found to be fictitious, most likely generated by an AI tool and included without verification. An independent expert panel, chaired by Wits University AI researcher Benjamin Rosman, was appointed to rebuild the document. “We are currently working with the experts that he appointed. That work is under way, and then we will be presenting something to him soon for consideration,” Morwane said.

Parliament has been clear about what it wants. Khusela Diko, chair of the portfolio committee on communications and digital technologies, told TechCentral that MPs want to strengthen existing regulators rather than create new ones. She pointed to bodies including Icasa, the Information Regulator, the Competition Commission and the National Consumer Commission as potential coordination points, and said the Information, Communication Technologies and Media Regulators Forum, launched in 2024, should serve as a crucial coordination vehicle.

“We don’t necessarily need to create a plethora of new institutions. Strengthen the institutions we currently have and be very specific on their mandates,” Diko said. “We are hoping for a policy that is perhaps a little more streamlined than the initial one, especially around the institutions.”

The protection of citizens from algorithmic discrimination sits at the heart of the debate. Diko identified a critical gap in how AI is currently treated: the reasoning process inside AI systems remains largely unexamined. “We are not effectively governing the reasoning process of AI systems: how AI connects the dots or how it gets from data to a decision,” she said. “The discriminatory laws of our past must not emerge in digital form.”

By contrast, the core principles of the original policy remain intact despite the withdrawal and redraft. The government still intends to ensure AI systems are trained in indigenous languages to prevent exclusion, and the revised policy will continue to focus on governance structures that mitigate AI risks while enabling the technology to support economic productivity.

South Africa has also signed the agreement establishing the World Artificial Intelligence Cooperation Organization, the China-led intergovernmental body established in Shanghai in mid-July. Morwane said the pact will “focus on creating native AI innovations and tools, rather than just using tools from other countries,” a commitment that speaks to the country’s ambition to shape AI development rather than simply consume it.

The revised policy is scheduled to reach cabinet in November, with January 2027 as a fallback. Broader public consultations will follow once the policy is finalized. “The actual timeline is we will introduce it to the cabinet in November. But considering all the holidays, we are just trying to be safe and say the latest will be January. So, it might happen sooner, depending on the cabinet schedule,” Morwane said.

Communications minister Solly Malatsi was scheduled to deliver a keynote address on South Africa’s evolving AI governance landscape at the Sandton symposium but tendered apologies and did not attend. Diko noted that the committee has had no further engagement with him on the redraft since the expert panel was announced, leaving open the question of how closely the minister will be aligned with the direction parliament and the department are now pursuing.