THE document was meant to chart South Africa’s course through the age of artificial intelligence. It went to the Cabinet. It was gazetted for public comment. Deputy President Paul Mashatile stood at a podium in Mpumalanga and spoke of it with ambition. Then someone read the footnotes.
What they found unravelled one of the most embarrassing episodes in South African policymaking in recent memory: the country’s Draft National Artificial Intelligence Policy – the document tasked with governing AI – had itself been written by AI, complete with a reference list populated by sources that do not exist.
By Sunday, Communications and Digital Technologies Minister Solly Malatsi had no choice but to pull the plug.
| “This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy.” — Minister Solly Malatsi |
“Following revelations that the Draft National Artificial Intelligence Policy published for public comment contains various fictitious sources in its reference list, we initiated internal questions, which have now confirmed that this was the case,” Malatsi said in a statement issued on Sunday.
“This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy. As such, I am withdrawing the Draft National Artificial Intelligence Policy.”
A POLICY WITH IMAGINARY SOURCES
The chronology of the debacle is damning in its detail. Cabinet approved the draft AI policy on 25 March 2026 – with a combined Special Sitting on 1 April, a date that will not be lost on satirists. On 10 April, it was published in the Government Gazette, and South Africans were given until 10 June to submit public comment.
At no point in that chain – from departmental drafting, to Cabinet consideration, to gazetting – did anyone appear to scrutinise the document’s academic scaffolding closely enough to notice that its citations were fabricated. The machinery of government processed a policy built on fiction.
The draft itself carried lofty ambitions. It extended South Africa’s initial AI policy framework with principles of intergenerational equity, promising that AI-driven innovation would prioritise the well-being of current and future generations. Mashatile, speaking at the launch of a Fourth Industrial Revolution lab and Centre of Specialisation in Mpumalanga earlier this month, said the policy would establish national priorities, norms, and sector-specific strategies across manufacturing, energy, infrastructure, transport, and trade.
None of that vision survives contact with the fundamental problem: the intellectual foundation on which the document rested was conjured, not researched.
| “This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical. It’s a lesson we take with humility.” — Malatsi |
CONSEQUENCE MANAGEMENT PROMISED
Malatsi did not attempt to minimise the gravity of what had occurred. His statement was marked by a tone of institutional remorse, and a pointed warning directed inward.
“The Department of Communications and Digital Technologies did not deliver on the standard that is acceptable for an institution entrusted with the role to lead South Africa’s digital policy environment,” he said. “The most plausible explanation is that AI-generated citations were included without proper verification.”
Then came the reckoning: “I want to reassure the country that we are treating this matter with the gravity it deserves. There will be consequence management for those responsible for drafting and quality assurance.”
In the language of the South African public service, consequence management means accountability proceedings – a world that has too often promised more than it delivered. Whether Malatsi’s warning translates into actual disciplinary action, or dissolves into the familiar bureaucratic half-light, remains to be seen.
THE IRONY THAT WRITES ITSELF
The spectacle carries an irony so perfectly constructed it might itself seem artificially generated. A government department charged with regulating artificial intelligence produced a policy document through the unregulated use of artificial intelligence — and submitted it to the public as an authoritative national strategy, without the most basic quality checks that any postgraduate student would be expected to apply.
The hallucination of sources – a well-documented failure mode of large language models, in which AI systems confidently cite papers, books, and authors that do not exist – is precisely the category of risk that AI governance frameworks are designed to address. South Africa’s draft AI policy managed to embody the very pathology it was drafted to manage.
It is, in a perverse sense, instructive. Malatsi acknowledged as much: “This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical. It’s a lesson we take with humility.”
The lesson, however, comes at cost. South Africa has lost months of policy development time. The credibility of a policy environment already under pressure has taken a blow. And on the global stage, where countries are racing to establish credible AI governance frameworks, the country now enters the next phase of that debate with egg firmly on its institutional face.
Minister Malatsi has offered no timeline for when a revised draft AI policy will emerge. What is clear is that when it does, the department will face a scrutiny it did not invite before – from civil society, from academics, and from a public now primed to check the footnotes.
South Africans deserve better, the Minister said on Sunday. On that single point, there is no dispute across any aisle.
