South Africa pushes AI policy to 2027 after AI hallucination scandal

South Africa’s national artificial intelligence policy has been postponed until January 2027. The delay occurred because the initial draft was retracted due to the inclusion of fictitious references, which some suspect were generated by AI. This incident is considered a significant and embarrassing setback for an African government’s technological aspirations.

The Department of Communications and Digital Technologies confirmed the revised timeline to a parliamentary committee on Tuesday, with acting deputy director-general Jeanette Morwane explaining that a revised draft will go to cabinet for approval by November 2026 before being released for public comment in January 2027. That puts the final policy roughly nine months behind its original schedule.

The original draft was released in April 2026 for public comment and was intended to position South Africa as a continental leader in AI innovation while addressing ethical and economic concerns. It was withdrawn shortly after a News24 report exposed that several of its citations were fictitious, references to nonexistent sources and documents that could not be verified.

AI researchers refer to this phenomenon as “hallucination”: large language models often generate information that sounds believable but is completely made up, especially when they lack sufficient knowledge.

Communications Minister Solly Malatsi admitted to Members of Parliament that internal reviews failed to identify issues before they were publicised. He called it a “massive oversight” and acknowledged a lack of transparency about whether and how AI tools were used to compile the document’s references. He maintained that the main content of the policy had not been seriously questioned, but the credibility of the entire document was damaged by what was discovered in its citations.

Two officials have since been placed on precautionary suspension pending a formal investigation. Director-General Nonkqubela Jordan-Dyani called the incident “highly regrettable” and said withdrawal was necessary to restore the credibility of the process.

Similar read: “Never trust, always verify”: South Africa’s costly lesson after AI hallucinations taint National AI Policy

Malatsi has now established a seven-member independent panel of experts to review the original draft, recommend revisions, remove flawed sections, and replace fabricated citations with verified references.

South Africa’s role in the broader conversation about AI in government

The irony at the centre of this story is difficult to miss. A government policy designed to guide how South Africa regulates and leverages artificial intelligence was itself compromised by the very problem AI critics have been warning about, the generation of convincing but false information by AI systems that governments and institutions are rushing to adopt.

The delay is particularly significant for South Africa because the country has been striving to be a leader in AI governance within the region, with other African nations closely observing its progress. In 2024, the African Union adopted a continental AI strategy, and many African countries are now developing their own national frameworks.

A poorly handled AI policy in South Africa would not only be a setback for Pretoria but also hinder the momentum in a region that requires trustworthy and strong AI governance to attract responsible investment and safeguard its citizens from the potential harms of inadequately regulated AI.

For South Africans, this situation has both practical and symbolic consequences. Practically, businesses and developers in South Africa have been awaiting clear regulations to understand what is allowed, what is required, and what protections are in place. This clarity will now be delayed by at least eight more months.

Symbolically, an official government document about AI that contains AI-generated falsehoods raises valid concerns about the thoroughness with which official institutions review work before implementing it as policy. It also questions whether the eagerness to appear technologically advanced is overshadowing the necessary discipline to ensure accuracy.

This episode serves as a cautionary tale for Nigeria and other African governments rushing to implement AI policies. While there’s pressure to develop frameworks demonstrating readiness for the digital economy, a policy built on false premises is more detrimental than no policy at all. It creates an illusion of governance while failing to address critical issues.