The irony cuts like a knife: South Africa’s first comprehensive AI policy just got torpedoed by artificial intelligence. The 86-page document, designed to establish regulatory frameworks for responsible AI use, was withdrawn after investigators discovered the policy itself contained six completely fabricated academic citations—courtesy of unverified AI-generated content. It’s like watching a fire department burn down while testing new sprinkler systems.
When AI Lies About Science
Generative AI tools created fake journals and non-existent research papers that sailed through multiple government review layers.
The mechanism behind this spectacular failure reveals how AI hallucinations actually work. When language models can’t find real sources, they confidently invent them—complete with authentic-sounding journal names and researcher credentials. Minister Solly Malatsi acknowledged the “most plausible explanation” was that AI tools drafted the policy and “AI-generated citations were included without proper verification.” Six phantom studies out of 67 total references made it through Cabinet approval and departmental reviews before News24 exposed the fraud.
Political Meltdown Meets Tech Reality
Opposition politicians called it “one of the most alarming failures of ministerial oversight” as the scandal echoes similar AI disasters worldwide.
The African National Congress Study Group didn’t mince words, condemning this as institutional failure rather than technical glitch. Their statement perfectly captured the absurdity: a policy meant to regulate AI was “itself produced through the uncritical and unverified use of Artificial Intelligence.” This pattern represents broader challenges facing institutions worldwide as they grapple with AI deployment without adequate oversight protocols.
The Costly Aftermath
The withdrawal delays South Africa’s entire AI governance framework by up to a year, leaving regulatory uncertainty during critical technology adoption.
Policy analysts now expect 6-12 months before a revised draft emerges, pushing legislative implementation potentially beyond 2027. The original policy proposed creating a National AI Commission, AI Ethics Board, and regulatory authority—structures now indefinitely postponed. As Minister Malatsi conceded, “This unacceptable lapse proves why vigilant human AI oversight over the use of artificial intelligence is critical.” The lesson writes itself.
