A regulator embarrassed by AI proposes to police AI
On May 21, 2026, Minister in the Presidency Khumbudzo Ntshavheni used the Government Communication and Information System's (GCIS) parliamentary budget vote to lay out a sweeping new anti-misinformation agenda: a forthcoming National AI Policy that would compel digital platforms to label AI-generated content, a prohibition on "broadcasting of Fake and Misleading News about South Africa," and a state-run fact-checking website empowering GCIS to "proactively debunk deepfakes and tactical misinformation" while inviting citizens to report suspected fake news directly to the department.
The timing is difficult to ignore. Just weeks earlier, on April 27, 2026, Communications Minister Solly Malatsi was forced to withdraw the government's own draft National AI Policy — the very document meant to govern this space — after journalists discovered that at least six of its 67 academic citations were AI-hallucinated: real-looking references to journals and authors that simply did not exist. Malatsi called it an "unacceptable lapse" and promised "consequence management." The policy had been Cabinet-approved on March 25 and gazetted for public comment on April 10; it lasted just 16 days before collapsing under scrutiny.
The steelman: South Africa's misinformation problem is real
GCIS's underlying concern is not manufactured. Ntshavheni told Parliament that South Africans rank among the world's most misinformed populations on the Ipsos Perils of Perception Index, and cited AI-generated deepfakes and synthetic news as a growing vector for eroding public trust ahead of a crowded municipal and provincial electoral calendar. A dedicated fact-checking resource that citizens can consult voluntarily is genuinely useful, and disclosure requirements for AI-generated political or news content have precedent — the EU's AI Act (Article 50) already mandates AI-content labelling for exactly this reason, and platforms like Meta and YouTube already self-label synthetic media. Transparency-by-labelling is a comparatively light-touch tool: it does not remove speech, it flags provenance and lets audiences judge for themselves.
Where the proposal overreaches
The trouble is that GCIS's framework does not stop at labelling. "Prohibiting the broadcasting of fake and misleading news about South Africa" is a content-based speech restriction, not a disclosure rule — and South Africa currently has no general statute criminalising "fake news" as such. Disinformation today is addressed narrowly and reactively: through defamation common law, the Electoral Code's rules around election-period falsehoods, and ICASA's broadcasting Code of Conduct, which is enforceable against licensed broadcasters through a Complaints and Compliance Committee but does not reach print, online, or social platforms, which remain under the voluntary Press Council. GCIS's proposal would fill that gap not with a narrowly tailored, judicially reviewable offence, but with an executive department — the government's own communications and PR arm — deciding, via a state fact-check platform, what counts as "misleading news about South Africa" and flagging it for takedown. That is a structural conflict of interest: the same ministry responsible for managing the government's public image would also adjudicate what is false about it.
This is precisely the failure mode South Africa just lived through. The National AI Policy scandal was not really about a rogue chatbot — it was about an institution outsourcing judgment to an unverified system and skipping the human review that would have caught six fake citations before Cabinet ever saw them. GCIS's own fact-checking platform introduces an identical single point of failure at much higher stakes: instead of vetting citations, government reviewers would be vetting the entire South African information ecosystem, without the transparency, appeal process, or judicial oversight that a defamation claim or an ICASA complaint currently affords.
What proportionate regulation looks like here
A disclosure-first version of this policy — mandatory AI-content labelling on major platforms, akin to the EU model — is defensible and consistent with global norms. A voluntary public fact-check resource that citizens can use to cross-reference claims is also fine, so long as it doesn't carry statutory takedown force. What should not survive drafting is a vague "fake and misleading news" broadcasting ban enforced by the same department that just proved, in public, that it cannot yet reliably distinguish real citations from AI-generated fabrications. If GCIS wants credibility on misinformation, the more urgent reform is fixing its own verification pipeline — not building a bigger one to police everyone else's.
Parliament's portfolio committee on communications should insist that any statutory "fake news" provision be narrowly defined, subject to independent judicial or Press Council-style review rather than departmental discretion, and paired with a published account of how the AI Policy's citation failure happened and what safeguards now exist. Until that accounting is public, handing GCIS adjudicatory power over "misleading news" asks South Africans to trust a verification process the government has not yet demonstrated it has itself.