South Africa's National AI Policy Draft Cited Sources That Did Not Exist

What happened

In April 2026, South Africa's Department of Communications and Digital Technologies gazetted its Draft National AI Policy for public comment, a step that invited researchers, civil society, and industry to scrutinize the document's proposals. Scrutiny quickly landed on something the drafters apparently had not noticed: the policy cited at least six academic references that independent reviewers could not locate, and journal editors in some cases directly disclaimed the papers the document had attributed to them.

Experts who examined the citations said the pattern matched a known failure mode of AI language models, the generation of plausible-looking but entirely fabricated sources. The journals, article titles, and authors carried the surface appearance of legitimate scholarship. The underlying papers did not exist. That gap between plausible form and missing substance is what makes AI hallucination in a formal government document different from a typo or a bad paraphrase. A hallucinated citation points to a source that no one can go back and check.

The Department of Communications and Digital Technologies acknowledged it was reviewing the discrepancies. It also argued that the fictitious references did not affect the draft's substance. That framing is difficult to sustain. A policy document uses citations to anchor its claims in established evidence. When the cited evidence does not exist, the factual basis for the policy's conclusions becomes unverifiable, not merely imperfect. Substance and citations are not separate things.

The subject matter sharpens the failure. South Africa was building a framework to govern how AI is developed and deployed inside its borders, including the responsible use of AI in government work. The draft illustrated the exact failure mode that AI governance frameworks are meant to address: outputs that look authoritative, released without verification, and carrying official weight simply by bearing a department's name. A gazetted policy document is not a draft memo. Other actors read it, cite it, and build positions on top of it.

The incident points to a structural problem that extends beyond this one document. No public record confirmed whether AI tools were used during drafting, which portions they generated, or whether any expert verified the citations against real sources before the document was gazetted. A provable record of what a system did, and what a human verified afterward, would make that gap visible before publication rather than after. Without it, any formal document produced partly by AI can enter official circulation with no indication of how much of its factual foundation is real.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.