South Africa's AI Policy Stalled by Fabricated Citations; Delivery Delays Expose Implement

South Africa’s draft national AI policy had to be withdrawn on 26 April after its 86-page text was found to contain fabricated academic citations. A redrafted version is now scheduled for cabinet consideration in November. That single operational failure captures, in miniature, the gap between AI’s projected economic benefits and the country’s current capacity to deliver them.

The promise is substantial, at least in some projections. Dawie Roodt, chief economist at Efficient Group, estimates that AI could add as much as two percentage points annually to South Africa’s GDP growth rate, on top of whatever baseline the economy achieves. His lower estimate of 1% reflects current conditions; the 2% assumes skills gaps and policy constraints are addressed. When asked whether the figure represented an annual boost or a one-time output increase, Roodt confirmed it as an ongoing annual addition.

That projection sits far above what other analysts consider realistic. PwC’s South Africa Economic Outlook, published in October 2025, models a “tense transition” scenario, which the firm describes as probably the most realistic future given regionalisation and nationalist fragmentation of the technology landscape. Under that scenario, South Africa would accumulate roughly 1.2 percentage points of AI-related benefit over the entire decade to 2035, or about 0.1 percentage points per year. Roodt’s estimate is more than 15 times larger.

The World Bank’s World Development Report 2026 falls between the two positions. Using methodology developed by MIT economist and Nobel laureate Daron Acemoglu, the bank modelled outcomes across 52 countries and found that advanced economies could see potential growth rise from 1.2% to 3.6% over the 2020s, while emerging markets could move from 4.1% to 4.9%, a gain of roughly 0.8 percentage points. Chief economist Indermit Gill characterised AI as a lifeline that developing economies “should grasp before it slips away,” noting that even worst-case scenarios would lift developing-economy potential growth above what he called “the dismal average of the first half of the 2020s.”

The backdrop for this debate is a productivity crisis. South Africa’s economy expanded just 1.1% in 2025, barely keeping pace with population growth estimated at 1.2% by Statistics South Africa. Growth did accelerate to 1.9% year-on-year in the first quarter of 2026, but the longer-term picture is bleak. Roodt characterises the shortfall as a fundamental failure of productivity growth, which under normal circumstances should contribute roughly 1% annually. “We are getting poorer on a per capita basis,” he noted, with most GDP growth driven by population expansion rather than efficiency gains.

Structural barriers compound the challenge. The country has high levels of social inequality and a large marginalised informal sector with minimal exposure to advanced technology. Only 2% of South African job vacancies in 2024 specified AI-related skills, representing approximately 17,000 positions out of 845,000 total openings. PwC South Africa chief economist Lullu Krugel framed the stakes plainly: “Like electricity in the 20th century, AI has the potential to create more jobs than it displaces if it is used to pioneer new forms of economic activity. But South Africa needs to use AI to the benefit of society as a whole, or risk leaving behind many people that have less access to technology.”

Meanwhile, adoption rates remain the single constraint on which all three analyses converge. About two-thirds of the productivity gap between rich and developing economies stems not from structural factors but from differences in how quickly new technology is taken up. By April 2025, nearly 25% of internet users in high-income countries had used ChatGPT, compared to just 5.8% in upper-middle-income countries like South Africa. The World Bank projects this adoption gap will widen from 24 percentage points to 46 percentage points by 2035. Gaurav Nayyar, who directed the report, warned that “the window to get this right is narrow.”

On the infrastructure side, the World Bank cited South Africa’s chronic load shedding, though recent improvements appear to have addressed this issue, as a reason hyperscale AI investment has flowed to Brazil and Malaysia instead. Local nominal GDP stands at approximately $427 billion against roughly $775 billion in projected 2026 capital spending by five American AI hyperscalers. That capital is not currently flowing here.

Roodt’s conditions for achieving his 2% annual gain centre on three operational requirements: skills development, infrastructure, and policy. He pointed specifically to delays in licensing satellite broadband, which have kept SpaceX’s Starlink out of South Africa, and to regulatory constraints around mining and black ownership requirements as policy obstacles. He also suggested that primary and secondary economic sectors would benefit most from AI adoption, though both the World Bank and PwC rank agriculture and manufacturing employment among the sectors least exposed to the technology.

Roodt acknowledged that his projections remain conditional. “Without AI, certain policy changes can lift GDP to 3% easily,” he said. “Add say 2% of AI to that, then a 5% GDP is certainly possible, even more.” Whether the redrafted AI strategy, due before cabinet in November, can begin closing that gap across skills, infrastructure, and regulation simultaneously is the question that will determine which of these forecasts looks prescient in a decade.