South Africa Scraps 2026 AI Policy After AI Invents 7 Fake Citations

South Africa Scraps 2026 AI Policy After AI Invents 7 Fake Citations

In a landmark reversal, South Africa has withdrawn its draft national AI policy after investigators uncovered that generative AI had fabricated seven fictitious citations — including non-existent reports from the South African Council for Science and Technology, the University of Cape Town’s Centre for Digital Governance, and a phantom African Union AI Ethics Framework. The policy, meant to guide ethical AI deployment across public and private sectors, was pulled from public consultation in April 2026 after editorial audits exposed systemic AI hallucinations.

How AI Hallucinations Undermined Policy Integrity

The most damaging hallucinations included fake legal precedents and fabricated academic studies. One citation claimed a 2024 report from a non-existent government body. Another referenced a non-existent study from Wits University. These weren’t minor errors — they were algorithmic fabrications masquerading as authoritative sources. Without human verification, AI-generated content erodes the very foundation of policy legitimacy.

Global Precedents in AI Governance

South Africa’s case is not isolated. In 2023, a New York lawyer faced sanctions after using ChatGPT to cite six fake court cases in a federal filing. In 2025, a European academic journal retracted a paper containing 14 AI-generated citations. But South Africa is the first nation to retract an entire national policy due to AI hallucinations — signaling a turning point in AI governance.

Why Public Sector AI Drafting Is Dangerous

Government officials admitted the draft was created using an unvetted commercial AI tool with minimal editorial oversight. This mirrors broader challenges: resource-constrained institutions increasingly rely on AI to compensate for capacity gaps. Yet as research from Academia.edu and WiredSpace shows, South Africa’s education system has long struggled with cultivating critical research habits. AI doesn’t fix this — it amplifies it.

Steps to Audit AI-Generated Policy Text

Experts from the University of Pretoria’s Department of Private Law urge immediate reforms:

• Require human verification of every citation in policy drafts
• Implement AI content watermarking and provenance tracking
• Ban unvetted LLMs in regulatory drafting until audit frameworks exist
• Train public servants in AI literacy and source validation
• Adopt standards aligned with the EU AI Act and OECD AI Principles

As South Africa recalibrates, the incident underscores a critical truth: AI can augment human judgment — but it cannot replace accountability. The withdrawal of this policy isn’t a failure of technology. It’s a warning. Without integrity, transparency, and critical oversight, even the best-intentioned policies risk becoming fiction.