Why Generic AI Is Failing South African Bid Teams — and What to Do About It
18 pages. Five verified incidents. Legal analysis. Practical recommendations for South African bid teams.
Download Free PDFNo sign-up required. Free to share and distribute.
Key Findings
Verified Incidents
A candidate attorney submitted a supplementary notice of appeal citing nine case authorities. Seven did not exist — generated by AI and submitted without verification. Judge Bezuidenhout held that relying on AI without verification is "irresponsible and unprofessional" and referred the matter to the Legal Practice Council.
Source: SAFLII — saflii.org/za/cases/ZAKZPHC/2025/2.html
Junior counsel admitted using an AI tool called "Legal Genius" — described as exclusively trained on South African legal judgments — which produced two fictitious case citations in heads of argument. Acting Judge Smit stated: "coherent and plausible outputs are not sufficient if they are false." Referred to the Legal Practice Council.
Source: SAFLII — saflii.org/za/cases/ZAGPJHC/2025/661.html
Minister Malatsi withdrew South Africa's Draft National Artificial Intelligence Policy after News24 found at least 6 of 67 academic citations were hallucinated. Editors of three academic journals confirmed the cited articles had never been published. The minister called it "an unacceptable lapse that compromised the integrity and credibility of the draft policy."
Source: SAnews.gov.za — Minister Malatsi official statement, 26 April 2026
Two senior officials were suspended after AI-generated references were found in a Cabinet-approved white paper. 102 of 148 references were unverifiable or fictitious. The department appointed two independent law firms and committed to reviewing all policy documents produced since 30 November 2022.
Source: Official DHA statement — gov.za, 30 April 2026
A Johannesburg High Court acting judge came under investigation for allegedly using AI to draft a judgment citing a non-existent case. The appellant's AI detection tool found the judgment carried "high risk in terms of citation integrity." Judgment in the related appeal was reserved.
Source: Legalbrief, 18 May 2026; News24, 16 May 2026
What Is in the Whitepaper
The full whitepaper goes beyond this summary — with detailed legal analysis, the complete incident timeline, a framework for evaluating AI tools in procurement contexts, and practical recommendations for South African bid teams.
Why administrative disqualification rates in South African government procurement are so high, and how AI hallucinations are making the problem worse for bid teams that have adopted generic AI tools.
An overview of the PFMA, MFMA, PPPFA, and B-BBEE requirements that govern South African government procurement — and why this framework is uniquely demanding compared to private-sector RFP processes.
A plain-language explanation of how large language models generate hallucinations, why statistical prediction is fundamentally different from document-grounded verification, and why tools trained on general internet content cannot reliably serve South African procurement.
Detailed analysis of all five documented AI hallucination incidents, including the Polish contractor case, court case citations, and the legal consequences that followed. Every incident verified against primary sources.
PRECCA Section 34A (effective 3 April 2024), the Register for Tender Defaulters, PFMA liability for false declarations, and the legal standard that "intent is not a defence" when unverified AI content is submitted in a legally binding document.
What a purpose-built AI tool for South African government procurement looks like — document-grounded extraction, confidence-gated review, immutable audit trails, and human-in-the-loop verification at every decision point.
Seven practical recommendations for bid managers and procurement consultants — from AI tool evaluation criteria to verification protocols and compliance vault management.
From the Courts and Government
"Relying on AI technologies when doing legal research is irresponsible and unprofessional."
Judge Bezuidenhout — Mavundla v MEC, KwaZulu-Natal High Court, January 2025
"Coherent and plausible outputs are not sufficient if they are false."
Acting Judge Smit — Northbound Processing v SA Diamond Regulator, Gauteng High Court, June 2025
"This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy."
Minister Solly Malatsi — withdrawing South Africa's National AI Policy, 26 April 2026
"It seems that these references were generated and attached to the document after the fact, as they are not cited in the body of the text."
Official DHA Statement — Department of Home Affairs, gov.za, 30 April 2026
Legal Framework
When you submit a government tender, you sign Standard Bidding Documents that are legal declarations. The consolidated SBD 4 (Bidder's Disclosure) covers declarations of interest, past SCM practices, and independent bid determination — all sworn to be accurate. An AI-generated false statement, even submitted unknowingly, can trigger consequences under multiple pieces of legislation.
Effective 3 April 2024. Companies can be held criminally liable for corrupt activities by associated persons if adequate prevention procedures were not in place. A company with no AI verification protocols may find Section 34A relevant if false content is submitted.
PRECCA establishes a national register that, following criminal conviction, restricts companies and their directors from public procurement for years. One hallucinated sentence in a signed declaration can start this process.
The Public Finance Management Act requires that all information in tender submissions is true and complete. Immediate disqualification applies to any submission found to contain false information — regardless of intent.
The Municipal Finance Management Act applies the same strict compliance doctrine to municipal procurement. The doctrine of strict compliance — established in South African case law — leaves no room for condonation of non-compliant bids.
18 pages. All incidents verified against primary sources. Legal analysis. Practical recommendations. Free to download and share.
⬇ Download Free PDFNo sign-up required · Free to share · Tender Aid, Cape Town · 2026