Communications Minister Solly Malatsi confirmed last week what many in South Africa’s digital policy community found almost too ironic to believe: the country’s draft national artificial intelligence policy had been withdrawn after internal reviews found it riddled with fabricated academic sources, apparently generated by AI itself. A framework designed to regulate artificial intelligence had become, instead, a demonstration of exactly what unregulated AI use can produce.
The discovery moved quickly through government ranks. Two officials involved in the policy’s development were suspended following the findings, and authorities established an expert panel tasked with rebuilding the framework from scratch. The speed of the response suggested the government understood the severity of the credibility damage.
That damage is real. South Africa has spent considerable effort positioning itself as a leading artificial intelligence hub within Africa, a center for responsible technological development and innovation. A policy document seeded with fictitious references cuts directly against that ambition. Experts have raised concerns that the scandal could undermine the country’s standing at precisely the moment it is trying to establish regional authority on AI governance.
The public reaction has been sharp. Social media platforms became immediate forums for citizens questioning how a document of such institutional weight could pass through review processes without anyone catching fabricated citations. The contradiction proved difficult to ignore: the very risks that AI regulation is meant to address had materialized inside the regulatory document itself.
By contrast, the broader lesson here is not unique to South Africa. Governments worldwide are racing to develop AI policy frameworks, often under pressure to move quickly and demonstrate technological fluency. The South African case illustrates what happens when that pressure overrides the basic verification steps that any serious policy document requires. Whether the fabricated sources resulted from AI tools used without adequate fact-checking, or from human reviewers who simply failed to scrutinize the citations, the outcome is the same: a public document that cannot be trusted.
The appointment of an outside expert panel signals an acknowledgment that internal procedures were insufficient. It is a reasonable corrective step (though it does not explain why those procedures failed in the first place). How the panel approaches its mandate, what verification safeguards it builds into the drafting process, and how transparently it communicates its progress will determine whether public confidence can be restored.
The withdrawal, while necessary, has also created a regulatory gap at an inconvenient moment. Artificial intelligence is advancing rapidly, and South Africa now faces the task of catching up on policy while managing the reputational fallout from this episode. The expert panel carries a dual burden: producing a technically sound framework and demonstrating that the government can handle technology responsibly.
The deeper question the episode raises is about institutional capacity across government more broadly. As more public bodies consider deploying AI tools to improve efficiency and reduce costs, the South African experience is a pointed reminder that technological adoption requires corresponding investment in oversight, verification, and human accountability. Speed without scrutiny is not efficiency. It is exposure.
Whether the new panel delivers a policy that earns genuine confidence, or whether this episode becomes a recurring reference point in debates about government competence and AI governance, remains to be seen.