South Africa just withdrew its national AI policy.
Because the AI policy had fake AI-generated sources in it.
Read that sentence until it stops feeling surreal.
Government officials wrote a document about regulating artificial intelligence.
Used AI to help write it.
AI hallucinated citations that don't exist.
Nobody noticed until Reuters did.
And now the policy is withdrawn.
This is the most perfectly constructed irony in the history of technology regulation.
But it's also a warning that goes far beyond South Africa.
Here's the problem this incident exposes.
AI hallucinations are confident.
They don't say "I made this up."
They produce citations with authors, journals, page numbers, and publication years.
They look like research.
They feel like research.
And if the human in the loop doesn't verify each one they become policy.
Now scale that problem.
How many government documents, regulatory filings, legal briefs, academic papers, and corporate memos have been written with AI assistance this year?
How many contained citations that were never checked?
South Africa caught theirs.
Others may not know yet.
Here's the deeper irony that makes this historic:
The world is racing to write AI regulation.
The Clarity Act. The EU AI Act. National AI strategies globally.
If the documents used to regulate AI contain AI-generated fictions
We're not regulating AI.
We're letting it regulate itself.
South Africa withdrew the policy.
The question is who else should.