195 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
11 new this week Library last updated 12 September 2026
← The incident library
HD-INC-180
Government · South Africa · 2026 · Fictitious and insufficiently verified citations in a government policy consultation document

South Africa withdrew its draft national AI policy after fictitious references were found in its evidence base

By Ellie Harris · Filed Draft gazetted 10 April 2026; fictitious references reported late April 2026; withdrawal announced 26 April 2026

Alleged: Department of Communications and Digital Technologies developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

South Africa withdrew its draft national AI policy after fictitious references were found in its evidence base

What happened

It was reported that South Africa’s Department of Communications and Digital Technologies published its Draft National Artificial Intelligence Policy on 10 April 2026. The Cabinet approved draft was open for public comment until 10 June.

Later that month, News24 reported that several academic sources cited in the draft did not exist. On 26 April, it was reported that the minister withdrew the draft after an internal process confirmed that the reference list contained fictitious sources.

The minister said the most plausible explanation was that AI generated citations had been included without proper verification. It was reported that he said the failure had compromised the integrity and credibility of the draft and that there would be consequences for those responsible for drafting and quality assurance.

The withdrawal was formally gazetted on 12 June 2026.

What an auditable version would have shown

This was a draft policy, not law, but it had been approved by Cabinet and released for public consultation.

The problem was relatively simple to check. The academic sources cited in the draft either existed or they did not.

The sources reviewed do not show what checks were carried out before publication or how the fictitious references made it through the drafting and quality assurance process.

For each citation, the department could have kept a record showing whether the source existed, whether it supported the claim being made, who checked it and what the check found.

Where the gap was

It was reported that the minister said AI generated citations included without proper verification were the most plausible explanation. That is the government’s stated explanation, not a published technical finding showing which tool or person produced each fictitious citation.

A CitationVerifier is designed to check whether a cited work exists and whether it supports the claim attached to it before publication. A ConductRecord keeps the citation, the check, its result and the decision that followed.

Neither determines whether the policy itself was sound.

What governance should have looked like

It was reported that the department withdrew the draft after its internal process confirmed fictitious sources, and that the minister said there would be consequences for those responsible for drafting and quality assurance.

Where a government document relies on cited sources, best practice would be to check every citation before publication. For important claims the record should also show who checked the source and whether it actually supported what the document said.

The issue is not whether government employees use generative AI. The issue here was that fictitious references made it through the process and into a Cabinet approved document released for public consultation.

Failure Pattern: it was reported that a Cabinet approved draft national AI policy was released for public consultation with fictitious sources in its reference list. The problem was identified after publication, and the department later confirmed that the references were fictitious.

Governance Principle: where an organisation publishes evidence based material, it should be able to show that the sources it relies on exist, support the claims being made and were checked before publication.

The reference implementation of CitationVerifier and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Department of Communications and Digital Technologies could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed August 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →