South Africa has withdrawn its flagship draft national artificial intelligence policy after an investigation confirmed that sections of the 86-page document contained fictitious academic citations likely generated by the very technology it was designed to regulate.
Communications and Digital Technologies Minister Solly Malatsi announced the withdrawal on Sunday, a day after News24 reported that at least six of the draft's 67 references either did not exist or could not be found in recognised academic databases.
Editors at the South African Journal of Philosophy, AI & Society and the Journal of Ethics and Social Philosophy each confirmed that the papers attributed to their publications had never been written or published.
"This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy," Malatsi said.
"The most plausible explanation is that AI-generated citations were included without proper verification. This should not have happened."
The draft had been approved by cabinet on 25 March, published in the Government Gazette on 10 April and opened for public comment with a deadline of 10 June.
It set out ambitious plans including the creation of five new oversight bodies: a National AI Commission, an AI Ethics Board, a dedicated regulatory authority, a National AI Research Institute and an AI Insurance Superfund modelled on the Road Accident Fund, designed to compensate citizens harmed by AI systems.
The policy also proposed tax breaks, grants and subsidies to encourage private sector collaboration on AI infrastructure, and aimed to position South Africa as a continental leader in responsible AI development.
All of those proposals are now in limbo.
Malatsi said there would be "consequence management" for those involved in drafting and signing off the document, though he did not name individuals or specify the accountability process.
"This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical," he wrote on X.
"It's a lesson we take with humility."
The episode is not an isolated incident.
A study published in the journal Nature found that the share of academic papers containing at least one potentially hallucinated citation rose from 0.3% in 2024 to more than 2.5% in 2025, amounting to over 110,000 papers with invalid references.
An analysis of papers presented at the NeurIPS 2025 machine learning conference found at least 100 fabricated citations scattered across more than 50 peer-reviewed works.
Courts have also been affected, with lawyers in multiple jurisdictions sanctioned or referred to professional bodies after submitting legal briefs containing AI-generated case citations that did not exist.
No timeline has been given for a revised draft, and the withdrawal leaves South Africa's AI governance framework in regulatory limbo at a moment when peers across the continent are advancing their own regulatory efforts.
The recap
- South Africa withdraws draft national artificial intelligence policy
- 6 of 67 academic citations were found to be fabricated
- Draft will be revised before being reissued for public comment
