Two weeks after South Africa released the draft of its National Artificial Intelligence Policy for public comment, a report by Article One has revealed that the document is riddled with fake AI-generated sources and references.
Recall that on April 11, TechMedia Africa reported that South Africa’s Department of Communications and Digital Technologies had published its Draft National Artificial Intelligence Policy for public comment.
The draft, as seen by TechMedia Africa, is an 86-page framework that, for the first time, explicitly names the country’s apartheid-era structural inequalities as a primary obstacle to responsible AI adoption.
However, the apparent comprehensive nature of the policy is now being called into question. Article One, in a letter to Minister of Communications and Digital Technologies, Solly Msimanga, said that AI-generated sources were used as references in the document.
The next day, News24 reported that it could exclusively reveal that at least six of the academic journal articles cited in the policy are completely fictitious.
Reactions trail the discovery
Msimanga, in a post on X, said that he has “asked the DG to investigate and take action against anyone found to be responsible for any wrongdoing.”
However, Khusela Diko, Chairperson of Parliament’s Portfolio Committee on Communications and Digital Technologies, in a post on X, advised the minister to withdraw the policy and “stop looking for a scapegoat (or should we say a scape-bot?).”
“With the integrity of the draft AI policy seriously called into question, here’s an alternative to counter-productive grandstanding. Withdraw it immediately so you don’t face further embarrassment and lose more credibility,” Diko wrote.
She further advised that an entirely new draft should be developed—without using ChatGPT—before being released to the public for comment.
Phumzile van Damme, Malatsi’s predecessor as DA national spokesperson, also expressed disappointment in a post on X.
“I should be angry, but I am disappointed. I was so excited. Happy. Planned to read it this weekend. I could cry from the frustration and disappointment,” she wrote.
Insisting that the draft must be withdrawn because it is an insult to the public expected to submit comments, the tech consultant and activist—who specialises in combating political disinformation and misinformation—said a junior staff member should not be blamed for the oversight.
“Part of the due diligence in the AI era is checking for an over-reliance on AI. Where was that? The Department and the Ministry failed to do this. Some junior staffer cannot be blamed for the lack of due diligence from both,” she added.
However, Msimanga found support from fellow DA cabinet member, Dean Macpherson, Minister of Public Works and Infrastructure, who described Diko’s remarks as “the very definition of grandstanding.”
The increasing use of AI with human oversight
The controversy adds to a growing list of incidents globally where AI tools have been used in high-stakes policy or research work without sufficient human verification.
- One notable example is the case involving Deloitte, which was forced to repay the Australian government after it was discovered that AI tools had been used in the preparation of a $440,000 report for the administration of Anthony Albanese. The report was later criticised for inaccuracies and lack of proper validation.
Similar concerns have also emerged in academia and legal practice, where AI-generated citations—often referred to as ‘hallucinations’—have appeared in court filings and research papers, raising serious questions about reliability and professional accountability.
The South African case shows the underlying challenge in the adoption of AI: while the technology can accelerate research and drafting processes, failure to implement rigorous human oversight can undermine credibility, erode public trust, and lead to costly reputational damage.