There is a profound irony haunting the halls of South African governance. We are currently using automated assistants to draft the very rules meant to regulate automation, only to have the machines “hallucinate” the foundation upon which those rules stand.
The crisis is twofold:
- Executive Failure: In April 2026, the Minister of Communications withdrew the Draft National AI Policy after discovering “phantom” academic sources.
- Judicial Crisis: The Gauteng High Court is currently grappling with legal practitioners presenting AI-fabricated case law as legitimate authority.
While AI tools evolve at a breakneck pace, these failures serve as a sobering reminder that professional duty remains an exclusively human burden.
The South African AI Policy Withdrawal: A Timeline of Fabricated Facts
On 26 April 2026, Minister Solly Malatsi took the unprecedented step of scrapping the Draft National Artificial Intelligence Policy. The 86-page document, intended to be a cornerstone for South Africa’s digital sovereignty, was built on “digital ghosts.”
The “Ghosts” in the Draft:
Investigation revealed that out of 67 references, at least six were pure hallucinations. These weren’t typos; they were entirely fabricated scholarship:
- Synthetic Citations: The draft cited a non-existent paper, “Müller Schmidt (2024)” in the European Journal of Law and Technology, to justify its regulatory sandboxes.
- Phantom Journals: Editors of the South African Journal of Philosophy confirmed that referenced articles cited to support the policy’s ethical framework simply did not exist.
Minister Malatsi noted: “This failure has compromised the integrity and credibility of the draft policy… It’s a lesson we take with humility.”
Courts vs. Chatbots: Why AI Hallucinations are a Career Risk
The judiciary’s confrontation with AI has been equally stark. In Northbound Processing (Pty) Ltd v The South African Diamond and Precious Metals Regulator (2025), a legal team used a tool named “Legal Genius” to submit heads of argument.
The result? Fictitious case citations. When the court asked for a correction, the legal team submitted a second version containing two additional fake citations.
The Legal Consequences:
- Mandatory Referral: Under Article 16(1) of the Code of Judicial Conduct, Acting Judge Smit was obliged to refer the practitioners to the Legal Practice Council (LPC) for gross incompetence.
- The “Sense-Check” Myth: The court ruled that time pressure does not excuse professional negligence. A “sense-check” is no substitute for methodical verification.
Why AI Hallucinates: Plausibility over Accuracy
To understand the “Scape-Bot” scandal, one must understand Large Language Models (LLMs). These are probabilistic engines. They do not “retrieve” facts; they predict the next word based on mathematical probability. They prioritize linguistic coherence over empirical truth.
The Professional Risks of Unchecked AI:
- Prejudice to Clients: Strategies based on non-existent law lead to catastrophic losses.
- Reputational Bankruptcy: A practitioner’s credibility is their currency; hallucinations are a form of professional insolvency.
- Erosion of Public Trust: When the state cannot verify its own policies, the regulatory framework loses its mandate.
Bridging the Governance Gap with Black Rocket AI
The “Scape-Bot” scandal highlights a systemic failure in how South African organisations adopt technology. The rush to automate has bypassed the essential frameworks of AI governance.
This is where Black Rocket AI intervenes. As specialists in technical hardening and AI governance, we help enterprises transition from “Shadow AI” risks to secure productivity.
- AI Exposure Audits: Our 14-day rapid audit identifies data leaks and hallucination risks before they reach a judge’s desk.
- AI Gateway: We provide the necessary guardrails and red-teaming to ensure your outputs are reliable, verified, and safe for public or legal submission.
A New Standard for “AI Readiness”
Moving forward, AI Readiness must be redefined. It is not about how well you “prompt,” but how rigorously you “audit.”
| Action Item | Professional Standard |
| Source Verification | Use verified, domain-specific databases (e.g., LexisNexis) over general LLMs. |
| Human-in-the-Loop | Treat every AI-generated citation as a “suspect” until cross-referenced. |
| Transparency | Only use AI tools that provide clear audit trails and cite original data sources. |
Conclusion: The Editor as the Final Guardian
The withdrawal of South Africa’s national policy proves that technology cannot be a “scape-bot” for human negligence. In a world of automated drafts, the professional’s value has shifted. You are no longer just a “writer” you are an interrogator and the final guardian of the truth.
Don’t let your organisation become the next headline. Secure your AI strategy with Black Rocket AI.
