PRETORIA: South Africa has withdrawn its Draft National Artificial Intelligence Policy after a government investigation confirmed that at least six of the document’s 67 academic citations were fabricated, most likely by the same AI tools the policy was designed to regulate. Minister of Communications and Digital Technologies Solly Malatsi announced the withdrawal on April 27, calling the discovery “an unacceptable lapse” that “compromised the integrity and credibility of the draft policy.” A new draft must now be written from scratch.
How Hallucinated Citations Brought Down South Africa’s AI Strategy
The draft policy had moved quickly through South Africa’s bureaucracy. Cabinet approved it on March 25, 2026, with a follow-up special Cabinet session on April 1. It was published in the Government Gazette on April 10 for a 60-day public comment window, with submissions due by June 10. The document framed AI governance around principles of intergenerational equity and proposed four new regulatory bodies to oversee artificial intelligence in the country: a National AI Commission, an AI Ethics Board, an AI Safety Institute, and an AI Insurance Superfund.
The collapse started with a News24 investigation published April 24. Reporters checked the policy’s reference list and found at least six citations pointing to academic articles that do not exist. Editors at three journals: the South African Journal of Philosophy, AI & Society, and the Journal of Ethics and Social Philosophy. Each independently confirmed to News24 that the articles attributed to their publications had never appeared there. The 67-item reference list was supposed to establish the scientific and academic foundation for the policy. A significant portion of it appears to have been generated by an AI tool and published without a single source being verified.
Malatsi confirmed that conclusion in his official statement published by SAnews.gov.za. “The most plausible explanation is that AI-generated citations were included without proper verification,” he said. “This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy.” He also committed to accountability: “There will be consequence management for those responsible for drafting and quality assurance.” No specific disciplinary action has yet been named.
What does South Africa’s AI policy scandal mean for founders using AI in high-stakes work?
It means the same failure mode can take down a pitch deck, a legal brief, or a due diligence memo. The problem is not that AI produced inaccurate output. That risk is well known. The problem is that no human checked. A government ministry published a policy with fabricated sources, and it cleared every internal review before anyone noticed.
A government ministry drafted, approved twice at Cabinet level, gazetted for public comment, and then had to retract a major policy document because nobody verified a single citation. If that chain of institutional review fails to catch AI hallucinations, so can a startup’s content workflow. The question is not whether your AI tools make mistakes. They do. The question is whether your process catches those mistakes before they reach a client, a regulator, or a public filing.
The irony is hard to overstate. South Africa used AI to help write the policy governing AI. The AI hallucinated the research that was supposed to justify the policy. It’s a documented institutional failure, on paper, with named officials and specific dates. For any founder who relies on AI for content that goes to clients, partners, or regulators, this is the clearest possible argument for mandatory human review before anything leaves the building.
There’s also a macro consequence worth tracking. South Africa was positioning itself to lead AI governance on the continent, a role that requires credibility with researchers, industries, and trading partners. That credibility is now damaged. The four proposed regulatory bodies remain hypothetical. And global AI sentiment has been shifting, with public skepticism of AI growing even as investment accelerates. South Africa’s setback removes another credible participant from the governance conversation.
What Comes Next for South Africa’s AI Governance
Malatsi has ordered a new draft. The document that Cabinet approved twice must now be rebuilt from scratch. No timeline has been announced. Consequence management has been promised but not specified. No suspensions, resignations, or formal disciplinary proceedings have been publicly named.
The revised policy will need to navigate the full approval path again: a new Cabinet sign-off, a Government Gazette publication, and a public comment period. Given the public scale of the embarrassment, expect the next draft to face significantly more scrutiny from independent researchers, civil society groups, and academic institutions that now have a direct stake in the outcome. That scrutiny, by Malatsi’s own admission, is exactly what should have happened the first time.
South Africa is not the only jurisdiction where AI regulation has hit unexpected obstacles. Maine became the first U.S. state to pass a moratorium on new AI data centers earlier this month, and federal AI legislation in the U.S. has repeatedly missed its own deadlines. For founders planning around regulatory clarity, the lesson from Pretoria is straightforward: government AI governance is moving slower and more unevenly than the technology itself. When frameworks do arrive, assume they’ll need revision. Build verification into your own processes before anyone outside your organization forces you to.
