The South African government has been thrust into an embarrassing spotlight after withdrawing its highly anticipated Draft National AI Policy due to the discovery of fabricated citations generated by artificial intelligence (AI). This blunder not only delayed the country's push towards becoming an AI leader on the continent but also led to the precautionary suspension of two senior officials in the Department of Communications and Digital Technologies (DCDT). The incident underscores the perils of unchecked AI use in critical policymaking processes, raising urgent questions about oversight, verification, and the integrity of official documents.

In a move that caught the nation by surprise, Communications Minister Solly Malatsi announced the policy's withdrawal on April 27, 2026, just weeks after it was gazetted for public comment on April 10. The 86-page document, intended to outline South Africa's vision for ethical AI development, regulation, and innovation, was found to contain at least six fictitious references. These included non-existent academic journal articles from real publications, a classic hallmark of AI 'hallucinations'—where tools like ChatGPT invent plausible but entirely fake information to fill gaps in their knowledge.

Timeline of the AI Policy Fiasco

The saga began with optimism. The DCDT released the draft policy to position South Africa as Africa's AI hub, proposing institutions like a National AI Commission, an AI Ethics Board, and an AI Regulatory Authority. It also suggested tax incentives, grants, and subsidies to foster private-sector collaboration in AI research and deployment.

However, alarm bells rang when civic group Article One and independent journalists scrutinized the bibliography. They uncovered references to phantom studies, such as articles in reputable journals like the 'Journal of Artificial Intelligence Research' that did not exist. One example cited a paper titled 'Ethical Implications of AI in Governance' by 'J. Smith et al.', which led to dead ends upon verification—no author, no DOI, no digital footprint.

By late April, the story broke in local media, prompting public outrage and expert commentary. Minister Malatsi acted swiftly, pulling the policy from consultation and initiating an internal probe. On May 1, DCDT Director-General Nonkqubela Jordan-Dyani suspended two officials pending the investigation's outcome.

The Suspensions: Accountability in Action

The suspensions mark a decisive response from the department. The two officials, described as senior policy drafters, were placed on precautionary leave with full pay while facts are established. This mirrors a parallel incident at the Department of Home Affairs, where two executives faced suspension over similar AI-generated inaccuracies in an immigration policy draft.

DCDT spokesperson Nonzwakazi Mkhize confirmed: 'An internal review process has been established to determine the full extent of the lapse and ensure it does not recur.' Minister Malatsi echoed this, stating, 'This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical. Consequences will follow for those responsible.'

While names remain undisclosed due to ongoing probes, sources indicate the officials relied heavily on generative AI without cross-checking outputs—a common pitfall in rushed document preparation.

Unpacking AI Hallucinations: How the Error Occurred

AI hallucinations occur when large language models (LLMs) confidently produce incorrect or invented information. In this case, the tool likely generated citations to bolster arguments on AI ethics and regulation, mimicking real academic style but fabricating details.

Experts like Tyronne McCrindle from Article One noted the references were 'plausibly fake,' citing real journals but imaginary papers. This isn't isolated; similar issues have plagued lawyers and academics globally, including South African courts where AI-cited cases led to sanctions.

The policy's reliance on unverified AI output bypassed standard research protocols, such as peer review or database searches via Google Scholar or JSTOR. As one analyst put it, 'AI is a powerful assistant, not a replacement for human diligence.'

Minister Malatsi's Response and Public Apology

Minister Malatsi addressed Parliament, calling the incident 'quite embarrassing' and a 'technical oversight.' He emphasized, 'This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy.' A new version is promised, with stricter guidelines on AI use.

The apology resonated with stakeholders, but critics argue it reveals deeper systemic flaws in government digitization efforts. Opposition parties demanded a full audit of AI-assisted documents across departments.

Photo by Emmanuel Ikwuegbu on Unsplash

Expert Reactions: A Wake-Up Call for Governance

AI ethicists and policy experts hailed the swift withdrawal but warned of broader risks. Dr. Sarah Gordon from the University of Cape Town noted, 'This highlights the epistemic integrity crisis: AI can undermine trust in authoritative sources if not governed properly.'

International observers, including Reuters, framed it as a cautionary tale for emerging economies racing to regulate AI. Locally, the incident snubbed South African AI experts who offered input but were ignored, per Article One.

Lessons include mandatory AI disclosure, verification workflows, and training for civil servants. The OECD AI Principles stress transparency and accountability—principles the draft ironically espoused but failed to embody.The Conversation analysis urges cross-sectoral cooperation to combat synthetic media harms.

What the Draft Policy Promised—and Why It Mattered

The document envisioned a comprehensive framework: risk-based regulation, data sovereignty, skills development via AI hubs, and international alignment with EU AI Act and African Union strategies. It targeted sectors like healthcare, agriculture, and finance for AI-driven growth, projecting R100 billion economic boost by 2030.

Stakeholders welcomed the ambition but criticized vague enforcement. The blunder delayed these goals, eroding momentum amid global AI competition.

Similar Blunders: Not Unique to South Africa

This echoes global cases: US lawyers fined for AI-faked cases, New York attorneys sanctioned. In SA, Home Affairs' immigration policy cited ghosts, prompting suspensions. Academics worldwide retract AI-polluted papers.

The pattern? Overreliance on tools without verification. As Reuters reports, South Africa's pullback signals rising AI governance needs.

Implications for South Africa's AI Ambitions

SA lags in AI readiness, ranking low in global indices. The scandal risks investor confidence but spotlights urgency. Positive spin: it accelerates robust policy with expert input.

Economically, AI could add 15% to GDP by 2030 per PwC, but requires ethical foundations. The incident boosts calls for national AI literacy programs and regulatory sandboxes.

Public and Industry Backlash

Social media erupted with memes dubbing it 'AI writes AI policy.' Tech firms like Microsoft SA urged caution, while startups fear delayed incentives.

Parliament's portfolio committee summoned Malatsi, demanding timelines. Civil society pushes for open consultations on the redo.TimesLive coverage highlights scrutiny.

Photo by Hennie Stander on Unsplash

Path Forward: Rebuilding Trust and Policy

DCDT pledges a revised draft by Q3 2026, incorporating stakeholder feedback and AI safeguards like watermarking and fact-check mandates. Training for 500 officials on ethical AI use is underway.

Globally aligned, it could position SA as a continental leader. Key: human-AI symbiosis, not replacement.

This blunder, while costly, offers a timely lesson. As AI permeates governance, South Africa must prioritize verification to harness its potential responsibly. The road to credible AI policy demands diligence, transparency, and collaboration—hallmarks absent in the first draft but essential for the future.