South Africa was forced to withdraw its draft National Artificial Intelligence Policy after it was discovered that the document contained fabricated academic citations, journals, and authors—a stark warning about the dangers of unsupervised AI use in government policymaking.
The Scandal: AI Hallucinations in National Policy
South Africa’s Department of Communications and Digital Technologies published a draft National Artificial Intelligence Policy for public comment in early 2026. The document appeared comprehensive, citing numerous academic sources to support its recommendations on AI governance, ethics, and economic development.
Then researchers noticed something was wrong.
Multiple citations referenced:
- Non-existent academic journals
- Fabricated author names
- Papers that were never published
- Institutions that don’t exist
The entire policy framework had been generated using AI tools without proper fact-checking or verification—a phenomenon known as “AI hallucination.”
Why This Matters
The South African AI policy scandal highlights several critical issues:
1. Regulatory Uncertainty
With the draft policy withdrawn, South Africa’s digital economy is left in regulatory limbo. Businesses seeking clarity on AI compliance, data governance, and ethical AI deployment now have no official guidance. This uncertainty could deter investment and slow AI adoption in a country that needs technological acceleration.
2. Credibility Damage
Government policies rely on expert authority and evidence-based reasoning. When fabricated citations are exposed, it undermines public trust in both the specific policy and the government’s capacity to regulate emerging technologies competently.
3. The AI Governance Paradox
South Africa attempted to regulate AI using AI-generated content without human oversight—the exact problem that AI governance frameworks are meant to prevent. The irony has not been lost on policy experts and technology ethicists.
Technical Details: How AI Hallucinations Occur
Large language models generate text by predicting likely word sequences based on training data. They don’t “know” facts in the human sense—they generate plausible-sounding content that may or may not correspond to reality.
When prompted to write a policy document with citations, AI systems can:
- Invent citations that follow academic formatting conventions
- Combine real author names with fake paper titles
- Create plausible journal names that don’t actually exist
- Reference real institutions with fabricated research outputs
Without expert human review, these hallucinations are difficult to detect, especially for non-specialists.
Global Context: AI in Government
South Africa’s scandal occurs as governments worldwide grapple with AI integration:
United States
- White House advocating for unified federal AI governance
- Multiple states passing comprehensive AI legislation (Connecticut, California, Colorado, Texas)
- AI litigation task force established to challenge inconsistent state laws
European Union
- EU AI Act officially in force since August 2024
- High-risk obligations applying from August 2026
- Strict requirements for transparency and risk management
United Kingdom
- Labour government moving toward structured AI regulation
- Departing from previous “pro-innovation” non-binding approach
Lessons for Policymakers
The South African incident offers critical lessons for governments developing AI policies:
- Human-in-the-loop is non-negotiable — AI can draft, but humans must verify every claim, citation, and recommendation
- Expert review is essential — Academic and industry experts should validate technical content before publication
- Transparency about AI use — Governments should disclose when AI tools are used in policy development
- Citation verification protocols — Implement mandatory fact-checking workflows for all referenced sources
- Version control and audit trails — Track all changes and contributions to policy documents
What This Means for AI Regulation
Paradoxically, the scandal demonstrates why AI regulation is necessary. Without standards for:
- AI system transparency
- Training data disclosure
- Output verification requirements
- Human oversight mandates
organizations will continue to deploy AI tools in high-stakes contexts without understanding their limitations.
South Africa’s Next Steps
The Department of Communications and Digital Technologies has:
- Withdrawn the draft policy pending complete revision
- Committed to developing a new framework with proper expert oversight
- Faced calls for accountability from opposition parties and civil society
The timeline for a revised policy remains uncertain, leaving South Africa’s AI sector in regulatory limbo.
Bottom Line
South Africa’s AI policy scandal is a cautionary tale for the AI age. As governments rush to regulate emerging technologies, they must resist the temptation to automate the regulatory process itself. AI is a powerful tool for drafting and analysis, but it cannot replace human expertise, judgment, and accountability. The countries that get AI regulation right will be those that use AI wisely—not those that outsource their thinking to it.
