South Africa Withdraws AI Policy Over Fake Citations

South Africa withdrew its Draft National Artificial Intelligence Policy 17 days after it was published in April 2026 because the document cited fake research generated by AI. The policy would have made South Africa the first African nation to establish a formal ethics board to oversee AI development outside the West.

At least six of the 67 sources listed in the bibliography were AI hallucinations, according to a letter from civil rights group Article One. A News24 article reported that several of the academic journals cited were completely fictitious. “The most plausible explanation is that AI-generated citations were included without proper verification,” South Africa’s minister of communications and digital technologies wrote in a statement.

AI-generated text or citations have entered official or quasi-official documents several times in the past two years. A Make America Healthy Again report on children’s health released in May 2025 by the Trump administration included incorrect citations.

The report listed nonexistent studies, mixed up author and journal attributions, and drew wrong conclusions from real studies. The Washington Post found that some references included “oaicite” attached to URLs, often considered a marker of ChatGPT use.

The White House press secretary described the errors as “formatting issues” and said a corrected report would be uploaded, which occurred hours later.

2025 the Australian Financial Review reported suspected AI use in a report commissioned by Australia’s Department of Employment and Workplace Relations from a company. Academics identified fake academic references and made-up quotes. The company confirmed that the use of generative AI tools had resulted in inaccurate outputs in citations and footnotes.

It republished the corrected study in September 2025 and refunded the government $290,000 of the $440,000 charged for the work. A separate 526-page healthcare report prepared for the Newfoundland and Labrador government in Canada by the same company and costing $1.2 million also contained fake citations, according to a report last November.

The company rereleased the report with corrected citations. The government updated its Request for Proposals contract to require disclosure of all intended uses of AI and machine learning and reserved the right to assess AI-related risks at any point.

Europe’s cybersecurity agency ENISA admitted that two of its threat reports published in 2025 contained AI-hallucinated sources. Out of 492 footnotes in one report, 26 were incorrect, according to researchers from the German public institution Westfälische Hochschule.

Researchers have expressed caution about assigning AI epistemic authority in official publications.