South Africa delays its National AI Policy to 2027 after discovering fabricated, AI-generated academic citations in the initial draft.
South Africa’s ambitious bid to position itself as the continent’s premier hub for artificial intelligence governance has hit an embarrassing roadblock. The Department of Communications and Digital Technologies (DCDT) officially announced that the timeline for the finalization of the National AI Policy has been pushed back to January 2027. The delay comes after the government was forced to abruptly withdraw an earlier draft when it was discovered to contain fabricated, potentially AI-generated academic references.
The policy, which was originally gazetted for a 60-day public comment period on April 10, 2026, was meant to mark a decisive shift from high-level ethical frameworks to formal, concrete sector-specific regulations. However, as first exposed by local investigative news publication News24, the 86-page document was riddled with “hallucinated” citations, academic sources, and journals that simply do not exist.
Appearing before a parliamentary committee, Communications Minister Solly Malatsi described the incident as a “massive oversight.” He acknowledged that internal quality checks failed to flag the errors and admitted to a severe lack of transparency regarding how generative AI tools were used by staff to compile the policy’s bibliography.
While Malatsi maintained that the core text and strategic pillars of the policy had not faced significant empirical challenge, the presence of fabricated data severely compromised the credibility of the entire process. In a swift bid to restore trust, Director-General Nonkqubela Jordan-Dyani confirmed that two department officials have been placed on precautionary suspension pending a formal internal investigation. Reports from CNBC Africa detailed that the withdrawal was an absolute necessity to prevent a wider institutional crisis.
To rebuild the framework from scratch, Minister Malatsi has appointed an independent, seven-member expert panel. This committee will thoroughly audit the original text, remove problematic parameters, and swap out the fabricated citations with fully verified, legitimate research.
According to Technext, the newly revised schedule outlined by acting deputy director-general Jeanette Morwane, the panel is expected to submit a clean, rectified draft to the Cabinet by November 2026. If approved, the document will be released fresh for public scrutiny and commentary in January 2027, putting the country nearly nine months behind its initial deployment schedule.
Before the scandal erupted, South Africa was moving toward a multi-regulator model to govern high-risk AI deployment, aiming to blend oversight across data protection, fintech, and digital infrastructure. Legal experts tracking the progression via Global Compliance News highlighted that the original framework was highly anticipated by businesses and developers needing clear regulatory boundaries to attract foreign investment.
The postponement leaves South Africa in a temporary regulatory vacuum at a critical juncture when global powers are moving aggressively to codify AI restrictions. Industry analysts warn that this misstep serves as a cautionary tale for neighboring nations within the African Union’s continental AI strategy, illustrating a profound irony: the very technology governments are rushing to regulate can undermine the policy-making process itself if left unchecked.
