The country was barely two weeks into the public comment window on its draft national AI policy. Then someone decided to check the citations.
That is how the South African draft AI policy fell apart last month.
Solly Malatsi is the country's digital tech minister. On April 26, he pulled the policy. The reason: at least six of its 67 cited studies do not exist.
The named journals say they never ran those papers. That list includes the South African Journal of Philosophy. It also includes the journal AI & Society.
The likely cause? An AI tool wrote the source list. The humans on the draft skipped the check.
The irony writes itself. The very plan to set national rules for AI got caught using AI without care.
The fake citations were "included without proper verification," Malatsi said in a statement.
He added that the slip shows why human checks on AI tools are key.
For investors with money in firms leaning hard on AI tools, this is the kind of risk playing out in public.
The team has promised action against the staff who drafted it.
Market Briefs breaks down stories like this into five minutes a day, with a free investing class for new readers.
Made-up AI facts are no longer just an idea.
It is the kind of slip that puts a top minister on the front page. It can force a whole country to start a policy over.
The cabinet had signed off on the draft on March 25. Public comment was set to run until June 10.
Now the draft is off the table. There is no new one. There is no date for one.
Every Big Tech earnings call this year has talked up AI as a growth story. Stories like this are the kind of risk that buyers and watchdogs will press on harder next time.
Firms paying for AI tools want to see real checks built in. Not just trust the bot and ship the work.
A model can write a clean draft in seconds. But it can also make up fake facts with the same calm tone.
The pulled draft was meant to set rules across many big parts of the country's economy. That list includes factories, energy, infrastructure, transport, and trade.
It will get rewritten. With humans checking the footnotes this time.
The bigger lesson lands far beyond South Africa. Any team using AI to draft real work needs a clear human step in the loop.
Without one, the bot's mistakes become the team's. And as this case shows, those mistakes can show up at the worst time.
For now, the cost is one botched draft. Next time, the cost could be a court case or a deal that falls apart. The lesson is the same either way.
Join Market Briefs for a daily read on what AI's rollout really looks like, with a 45-minute investing course as a bonus when you sign up.
