Africa tech

South Africa's AI policy and the fake references: a warning for every office using AI

Chrisphine Otieno · 19 Oct 2026 · 2 min read

South Africa's AI policy and the fake references: a warning for every office using AI

There is a particular irony in a national artificial intelligence policy undone by an artificial intelligence failure. South Africa's draft national AI policy was withdrawn after fictitious academic references were found in it, and disciplinary proceedings are continuing against one official, TechCabal reported on 15 October. Communications Minister Solly Malatsi has said the matter cannot be allowed to slide, has appointed a seven-member expert panel to review the document, and aims to release a revised policy for public comment by January 2027.

The episode drew attention across the continent, partly because it was embarrassing, and partly because almost every office now uses AI to draft documents.

What "hallucination" means

AI language models generate text that sounds right. Most of the time it is right. Sometimes it is not, and the model gives no warning. A well-known failure is invented citations: references to papers, authors, cases or reports that look real but do not exist. Lawyers in several countries have been sanctioned for submitting court filings with fabricated cases. Now a national policy draft has joined the list.

Why it keeps happening

  • AI writes fluently, so errors are hard to spot.
  • Citations look authoritative, so readers trust them.
  • Busy people skim, especially under deadlines.
  • Nobody owns the checking, when a document passes through many hands.

A checking routine for any office

Whether you are a ministry, a law firm, an NGO or a marketing team:

  1. Treat AI output as a draft from a confident intern. Useful, fast and unverified.
  2. Check every fact, figure, name and reference against its source. If you cannot find the source, remove the claim.
  3. Never let AI invent sources. Give it your sources and ask it to work from them.
  4. Name a person responsible for verification on every document.
  5. Keep a record of what was AI-assisted.

The same rule for marketing

This applies to adverts too. An AI-written script can contain a wrong price, an invented statistic or a claim your product cannot back up. In Tari Studio, scripts written with /script arrive as editable drafts; we ask you to check every word, especially prices, phone numbers, health claims and anything that sounds like a statistic, before you make the video. Scripts with /script.

The blog you are reading follows the same rule: news posts link to their sources so you can check them.

A warning, not a reason to stop

The lesson from South Africa is not that governments should stop using AI. It is that AI changes where human effort is needed: less on drafting, more on checking. Offices that adjust will gain speed without losing credibility. Those that do not will learn the lesson in public.

Read: truth in advertising when anything can be generated.