Ghost Authorities: AI Hallucinations In The Courtroom

In Molawa, Springkaan, Smith v Matjhabeng Local Municipality and Another,(“Molawa”) Judge Allen-Yaman J was confronted with a disciplinary chairperson’s ruling that cited case authorities which, upon scrutiny, simply did not exist.

The Molawa judgment is the latest, and among the most consequential,in a growing line of authority addressing the submission of AI-generated, fictitious case citations to courts and tribunals. The Molawa Judgment is a pointed reminder that the convenience of artificial intelligence cannot substitute for the irreplaceable duty of independent legal reasoning.

Background and procedural history

The applicants, three municipal employees of the Matjhabeng Local Municipality, were charged with misconduct on 12 March 2021. The disciplinary hearing was delayed by interlocutory disputes, ultimately resolved by agreement in review proceedings (Case No. JR1790/21) on 9 October 2024. Despite this resolution, the municipality took no further steps to prosecute the disciplinary charges until 2025.

When the hearing was eventually reconvened, the applicants raised a point in limine before the appointed chairperson, Dr Dan Raseluma, arguing that the municipality had waived its right to discipline them by virtue of its non-compliance with the time periods prescribed in the South African Local Government Bargaining Council Disciplinary Procedure Collective Agreement.

The chairperson dismissed the waiver argument in a written ruling dated 3 June 2025. An application for his recusal was similarly refused on 21 June 2025. The applicants then launched a review application (Case No. 2026-152359) challenging both rulings.

When the municipality refused either to stay the reconvened disciplinary hearing or to appoint a new chairperson, the applicants brought an urgent application to the Labour Court for a stay of the disciplinary proceedings pending the outcome of the review.

The phantom authorities

At the heart of the urgent application was a startling allegation: that the chairperson’s ruling on waiver relied on fabricated case authorities.

Of four cited authorities, two were entirely fictitious, and a third was materially misrepresented. Only one was both genuine and accurately characterised. The respondent’s argument that the chairperson “must have had other authorities in mind” was rejected. Allen-Yaman J observed:

“The proposition that any reader of a decision or judgment ought to infer from the surrounding discussion… that the decision maker intended to refer to a judgment other than the one clearly expressed merely needs to be stated to be rejected.”

Court’s findings

The court acknowledged that the precise source of the erroneous citations had not been conclusively determined. It held, however, that either explanation led to the same result: if the chairperson relied blindly on an AI tool that generated non-existent authorities, the ruling could not stand; equally, if the errors were self-generated and the chairperson failed to explain or correct them when afforded the opportunity (through the recusal application), this constituted prima facie evidence of a failure to apply an independent mind to the law.

Allen-Yaman J held that:

“If the Chairperson did, indeed, blindly rely on an AI ‘assistant’, South African courts and others have already expressed the unacceptable nature of such conduct”

The court found that exceptional circumstances justified intervention in the incomplete disciplinary proceedings. Continuing a process chaired by an individual whose decision-making and reasoning had been called into serious question would risk irreparable harm and grave injustice to both parties, particularly if the rulings were subsequently set aside on review.

A growing body of comparative authority

The Molawa Judgment situates itself within an increasingly robust body of domestic and international authority addressing the dangers of AI-generated legal citations.

In Parker v Forsyth NO and Others ,the court admonished practitioners in measured but firm terms:

“In this age of instant gratification, this incident serves as a timely reminder to, at least, the lawyers involved in this matter that when it comes to legal research, the efficiency of modern technology still needs to be infused with a dose of good old-fashioned independent reading. Courts expect lawyers to bring a legally-independent and questioning mind to bear on, especially, novel legal matters, and certainly not to merely repeat in parrot-fashion, the unverified research of a chatbot.”

In both Mavundla v MEC Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal and Others,(“Mavundla”)and in Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator and Others (“Northbound Processing”),the courts referred the issues to the Legal Practice Council for investigation,signalling that reliance on AI-hallucinated authorities may cross the threshold from carelessness into professional misconduct.

The English High Court in Ayinde v The London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC described the risk with characteristic precision:

“In the context of legal research, the risks of using artificial intelligence are now well known. Such tools can produce apparently coherent and plausible responses to prompts, but those coherent and plausible responses may turn out to be entirely incorrect. The responses may make confident assertions that are simply untrue. They may cite sources that do not exist. They may purport to quote passages from a genuine source that do not appear in that source.”

Most significantly, the Supreme Court of India in Singh v Jammu and Kashmir Bank Ltd and Another held that the inevitable consequence of the use of “non-existent, fake and hallucinated material” in support of a judgment is the setting aside of that judgment. This represents the most far-reaching judicial consequence to date: the integrity of the adjudicatory process itself is compromised when its reasoning rests on fabricated foundations.

Practical implications for legal practitioners

The Molawa judgment, read together with the comparative authorities, yields the following practical imperatives for legal practitioners:

• Verify every authority independently. AI tools are capable of generating citations that are superficially plausible, complete with case names, neutral citations, and law report references, yet refer to judgments that have never been delivered. Every authority produced by an AI tool must be independently verified against official law reports or approved electronic databases before inclusion in any submission, opinion, or ruling.
• Apply an independent mind. The duty is not merely to cite accurately, but to read, understand, and independently evaluate the authority’s relevance and correctness. A practitioner who submits an AI-generated list of authorities without reading the underlying judgments has abdicated the very function for which the profession exists.
• Understand the disciplinary exposure. The referrals to the Legal Practice Council in Mavundla and Northbound Processing signal clearly that the submission of fabricated authorities is not treated as a mere oversight. Practitioners face potential investigation and disciplinary proceedings that could affect their right to practise.

Conclusion

The Molawa Judgment stands as a salutary warning. AI is a tool of extraordinary potential, but it is precisely that: a tool. It does not reason; it generates. It does not verify; it predicts.

The legal profession’s commitment to accuracy, integrity and independent reasoning cannot be outsourced to an algorithm. As the Molawa judgment makes clear, the consequences of doing so may include the stay or setting aside of proceedings, referral to the Legal Practice Council, and the irreparable compromise of a decision-maker’s credibility.

Practitioners would do well to heed the admonition in Parker v Forsyth: the efficiency of modern technology must still be infused with “a dose of good old-fashioned independent reading.”

In an age where AI can fabricate with alarming plausibility, that old-fashioned virtue has never been more indispensable.