South Africa Used AI to Write Its AI Policy. The AI Lied.

In one of the most perfectly absurd stories to come out of African tech governance this year, South Africa had to withdraw its national AI policy, because the document meant to govern artificial intelligence was itself written carelessly using artificial intelligence.

The Draft National Artificial Intelligence Policy had been approved by Cabinet, published in the Government Gazette, and was open for public submissions. Then researchers started reading the footnotes. At least six of 67 academic citations in the document were fake. The journals cited do not exist. The articles cited were never published. Editors of the referenced publications independently confirmed to local media: these papers are not real.

Two officials have been suspended. The policy has been pushed to 2027.

Major Highlights

• South Africa’s Draft National AI Policy was published in the Government Gazette on 10 April 2026, with public submissions open until 10 June 2026.
• At least 6 of 67 academic citations were AI-generated hallucinations:  Journals, articles, and authors that do not exist in the real world.
• Editors of the South African Journal of Philosophy, AI & Society, and the Journal of Ethics and Social Philosophy all confirmed to News24: the cited papers had never appeared in their publications.
• The policy had cleared Cabinet approval on March 25 and April 1: Two rounds of government were signed off with no citation verification.
• Communications and Digital Technologies Minister Solly Malatsi withdrew the policy after the scandal broke.

  

• Two departmental officials placed on precautionary suspension pending formal investigation.
• The policy proposed five new oversight bodies including a National AI Commission and an AI Insurance Superfund. These are ambitious proposals that now have to wait until 2027.

KINI BIG DEAL

This story is not really about South Africa getting caught. It is about what happens when AI is used lazily.

The officials who drafted this policy probably used AI to speed up their research — a reasonable instinct. The problem is they published the output without verifying a single citation. They trusted the tool more than they verified the work. And the tool, as AI tools sometimes do when pushed beyond what they actually know, filled in the gaps with confident-sounding fabrications.

This is why this story reaches far beyond Pretoria. Every professional using AI to research, write, or support decisions needs to understand that AI is a first-draft colleague, not a final authority. The moment you hand over your responsibility to verify, regardless of what you are working on, you are the one who takes the fall when it is wrong.

AI go help you. But always check the receipts.

Read more: CNBC Africa | The Next Web | Technext

Until next time, stay curious.