In April South Africa made international news, but not in a good way.
Usually we make the international news because of politics or sport. Like the 2024 election, or the charges against Israel at the International Court of Justice. Or some celebrity is involved, like Elon Musk in the Starlink case.
Eggs, errors and eish
This time the SA government got enough egg on its face to make a Soweto-sized omelette. The Department of Communications and Digital Technologies proudly issued its Draft National AI Policy. And then quickly withdrew it. Why? The document had been generated using AI, and contained fictitious references.
Maybe we should call it EI, for Eish Intelligence.
I skimmed through the 86-page draft policy. It reads like a political speech: lots of buzz words like “strategic imperative”, but very little substance.
Another eggy eish moment
Zapiro produced the perfect cartoon for this blunder. A robot is spitting out a thick document while a group of government officials at a table are eating, gambling and sleeping. The scene reminded me of some trips to the Department of Home Affairs (DHA), although any robot there would also have been on lunch.
Speaking of Home Affairs, fake AI references also appeared in the Revised White Paper on Citizenship, Immigration and Refugee Protection. More egg, this time on the face of DHA. I wonder what other official documents have similar flaws. Perhaps Eskom uses AI for its reports about grid stability.
Past broken eggs
If you’ve only used AI to spruce up your email, you might not have experienced its hallucinations. It’s not just fake references, which are easy to check. AI can produce beautifully written nonsense.
Perhaps the group responsible for the policy hadn’t spent much time researching AI. But how did they miss the cases that hit the limelight? Two scandals immediately come to mind:
- In 2025, Deloitte had to refund part the Australian government part of the fee for an “expert” report. The report had contained serious AI-generated inaccuracies and fake references.
- Closer to home, there was a court case right here in Johannesburg. Lawyers had used AI to generate legal documents that contained non-existent citations. After this came to light, the lawyers submitted a revised version. The corrected version contained more fake citations.
Paying for the omelette
This is (unintentionally) the most accurate demonstration of AI governance challenges we could ask for. The story of the withdrawal can be included in the next version of the policy as Appendix A: “Why human oversight matters.”
The legal and ethical responsibility for AI actions is a huge issue at the moment. There have been court cases about AI companions that encouraged suicide or homicide. The next may be AI tools that dispense medical advice. The focus has been on holding AI developers responsible for inadequate safety measures.
It’s not just AI developers. AI users should beware. There is a warning in small font at the bottom of the ChatGPT screen: “ChatGPT can make mistakes. Check important info.” Microsoft includes this: “Copilot is an AI and may make mistakes. Using Copilot means you agree to the Terms of Use.”
The Joburg court case already makes it clear that you are responsible for your work. Not checking AI content in your work is irresponsible and lazy.
The good news (maybe not for them) is that unnamed officials were suspended for both the AI policy and DHA policy mistakes. This being South Africa, however, they have probably been suspended on full pay. And by the time the investigation is complete, they may have had months of paid holiday.
What do you think?