If a ministry’s AI policy is written by AI, read the footnotes first
South Africa withdrew a white paper full of invented citations. Kenya is trying to legislate before that embarrassment. Both are ahead of doing nothing.
Illustration. Reporting is based on company statements and published accounts of the round or the bill.
South Africa’s communications ministry had to pull an artificial-intelligence white paper because the document had been produced with AI and the citations did not exist. It is a comic fact. It is also the cleanest argument for the policy itself.
A tool that can imitate a government memo, including the bibliography, is already in every serious office on the continent. Pretoria found out in public. Most institutions will find out in a contract, a court, or a budget note that nobody checks.
Kenya’s Senate bill takes the other route: name an official, give that office the power to inspect and fine, and leave a sandbox so a startup is not treated like a bank on day one. The bill can still be written badly. A commissioner with inspection powers can harass companies. A sandbox can become a waiting room. Those are arguments about drafting, not about whether the subject is real.
The AfCFTA secretariat’s estimate, more than $700 billion in digital-economy value by 2035, will be quoted too often and examined too little. The part worth keeping is Mene’s narrower point. Trade in software needs rules that investors can read. A protocol on digital trade is one. A national AI law that a lawyer can cite is another.
Withdraw the bad draft. Publish the sources. Then write the rule. That sequence is available to every capital that just watched Pretoria do the first two steps the hard way.