South Africa Unveils Sweeping Draft AI Policy With Seven New Oversight Bodies
Communications Minister Solly Malatsi has officially gazetted South Africa’s draft national artificial intelligence policy, opening a 60-day public comment window on an ambitious 86-page document that lays out a bold — and some might say bureaucratically heavy — framework for governing AI across the country’s economy.
The draft policy, published in the Government Gazette and approved by cabinet on 25 March, is built around six strategic pillars covering education, digital infrastructure, data governance, ethics, and public sector deployment. It proposes dozens of targeted interventions across these areas, signalling that Pretoria is taking the AI governance challenge seriously — perhaps more seriously than its institutional capacity might currently allow.
South Africa’s Draft AI Policy Proposes Massive Regulatory Framework
The most eye-catching aspect of the document is its sheer institutional ambition. The policy puts forward the creation of seven new oversight bodies, including a National AI Commission to lead policy coordination, an AI Ethics Board to tackle issues of bias and privacy, and an AI Regulatory Authority tasked with conducting audits and issuing certifications.
Beyond these, the policy proposes an AI Ombudsperson Office that would allow ordinary citizens to challenge AI-driven decisions that affect them. A National AI Safety Institute and an Integrated AI-Powered Monitoring Centre are also on the table. Perhaps the most unusual addition is an AI Insurance Superfund, modelled directly on the Road Accident Fund, designed to compensate individuals or organisations harmed by AI systems in cases where liability is difficult to pin down.
The policy also envisions a repositioned role for Icasa, the country’s communications regulator, expanding its mandate to oversee ethical AI use across telecommunications, ICT, and broadcasting. A new National AI Regulatory Forum, coordinated by the Department of Communications and Digital Technologies, would bring together Icasa, the Information Regulator, the Competition Commission, the South African Reserve Bank, the Financial Sector Conduct Authority, the CSIR, and the Department of Trade, Industry and Competition.
The scale of this proposal naturally raises hard questions about funding and capacity, especially given the government’s well-documented struggles to adequately resource existing institutions. The draft does not attach specific budget figures to any of the proposed new bodies, although it does indicate that funding would be secured during the second year of a planned three-year implementation road map.
To its credit, the Department of Communications has been transparent about where this document stands. An explanatory note published alongside the policy openly describes it as “a work in progress”, noting that the final approach will require extensive consultation with both local and international experts. The draft is intended as “a point of departure and indication of government’s current thinking,” rather than a fixed final position.
On the regulatory philosophy, the document doesn’t commit to a single approach. It presents four broad models — an ethics-first framework, a flexible iterative model using regulatory sandboxes, an economy-focused strategy, and alignment with global standards — suggesting that a tailored combination of all four would be the ideal path forward. Several additional options are floated, including principles-based regulation, a guardrails model, and a “just AI” framework aimed at addressing inequality.
The policy builds on a national AI framework published in August 2024, drawing on 32 stakeholder submissions and cabinet cluster consultations. Education, healthcare, and agriculture are identified as priority sectors, with public administration serving as a critical implementation lever. The document calls for AI to be embedded in school curricula from primary through tertiary level and for community-based AI education centres to be established in underserved areas.
On the infrastructure side, the policy pushes for investment in supercomputing facilities, 5G and 6G networks, high-capacity fibre, and low-Earth-orbit satellite connectivity to reach remote communities. It goes a step further by proposing that universal internet access be recognised as a socioeconomic right and calling for the creation of “regional AI factories” — decentralised compute hubs meant to support local data sovereignty and stimulate regional economic growth.
The ethical dimension of the framework is grounded firmly in South Africa’s Constitution and Bill of Rights. The policy lists specific constitutional provisions that AI must not be used to violate. Notably, it invokes the African philosophy of ubuntu — emphasising interdependence, community, and shared responsibility — as a guiding lens for how AI should be developed and deployed in the South African context.
Proposed Safeguards
Among the key safeguards proposed are mandatory human rights and gender impact assessments for AI systems operating in high-risk domains, as well as requirements for human-in-the-loop oversight in critical decision-making environments. Public sector and high-risk AI systems would also be required to meet standards of “sufficient explainability” and “sufficient transparency.” Additional protections are proposed for children, specifically guarding against manipulative AI-driven advertising and gamified features designed to maximise screen time.
The implementation plan is structured in three phases. In the current 2025/2026 financial year, the policy would be finalised and draft regulatory requirements for “unacceptable risk” AI uses would be identified. Year two (2026/2027) would see those guidelines published and high-risk regulatory requirements implemented across sectors. Full implementation is targeted for 2027/2028, with the entire framework subject to a comprehensive review every three years — or sooner if major technological or legislative shifts demand it.
Written submissions on the draft policy are due by 10 June 2026. Whether the ambitious institutional blueprint survives contact with South Africa’s fiscal realities and the public consultation process remains to be seen, but the document marks a significant step in the country’s attempt to shape its AI future on its own terms.