South Africa published its draft National AI Policy for public comment on 10 April 2026, then withdrew it 16 days later after journalists found at least 6 of 67 citations were AI-hallucinated fake or misattributed studies.
67 citations
Academic citations in the draft policy (April 2026)
6 citations
Citations found fabricated or misattributed (April 2026)
16 days
Days between publication and withdrawal (10-26 April 2026)
Details
Maturity
Discontinued
Promoter
Department of Communications and Digital Technologies (South Africa)
Period
April 2026
Keywords
AI policy, public communications, digital governance
Context
On 10 April 2026, South Africa's Department of Communications and Digital Technologies (DCDT) published its Draft National AI Policy for public comment.
Activities
Weeks later, journalists and the civil-rights group Article One found that at least 6 of the document's 67 academic citations were fabricated: some referenced journals that do not exist, others cited real journals for articles that were never published there. Analysts concluded the drafters had used a generative-AI tool to help write the policy — the very technology the document was meant to govern — without verifying a single citation.
Results
Communications Minister Solly Malatsi withdrew the draft on 26 April 2026, 16 days after publication, calling the lapse 'unacceptable' and promising 'consequence management' for those responsible.
Conclusions
South Africa's case became one of the first known instances of a government withdrawing an official document specifically because of AI hallucinations, and was quickly cited alongside similar failures (Europe's ENISA cybersecurity agency later admitted AI-hallucinated sources in two 2025 threat reports) as evidence that public bodies drafting AI-related policy need mandatory citation verification and disclosure of AI-assisted drafting before publication.
Implementation
Indicative cost
Low (< €50k) — Not quantified; the direct cost was the drafting and withdrawal of a single policy document, though the reputational cost was significant.
Time to results
Short (< 1 year) — Published 10 April 2026; citation problems exposed within weeks; withdrawn 26 April 2026 — a 16-day lifespan for the draft.
Staffing & skills
Department of Communications and Digital Technologies (DCDT) drafting team produced the policy, Communications Minister Solly Malatsi made the public withdrawal announcement and ordered a 'consequence management' process
Conditions for success
Mandatory verification of every citation before publication
Disclosure requirement when a document has been AI-assisted in drafting
A named consequence-management process once the failure was found, to preserve institutional accountability
Common failure modes
Generative-AI tool used to help draft the policy without any citation verification, producing at least 6 of 67 fabricated or misattributed academic references
No disclosure of AI-assisted drafting was made before publication
The failure surfaced only after external journalists and a civil-rights group (Article One) audited the citations post-publication, not through an internal pre-publication check
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
