A South African Labour Court has halted disciplinary proceedings against three municipal employees after finding prima facie evidence that the chairperson of the hearing may have relied on hallucinated artificial intelligence (AI)-generated legal authorities in key rulings, raising serious questions about the integrity of the disciplinary process.
In a judgement handed down this month, Judge K Allen-Yaman stayed the disciplinary hearing involving employees of the Matjhabeng Local Municipality pending the outcome of a review application challenging two rulings made by the chairperson.
The employees had been charged for misconduct, with a disciplinary hearing to follow after the allegations were made.
At the heart of the dispute is the alleged use of legal authorities that either do not exist or were inaccurately cited – a growing concern as AI tools become more widely used in legal research.
Last year, ITWeb reported that a local judge ordered an investigation into the conduct of lawyers who used generative AI to present non-existent citations in court.
In the latest case, the applicants argued that apart from one correctly cited case, the authorities relied upon by the chairperson in a June ruling either did not exist, or did not support the legal propositions for which they were cited.
They contended that the citations were likely generated using AI and that this cast serious doubt on whether the chairperson had independently applied his mind to the matter.
The court agreed that several cited judgments could not be verified.
The municipality argued that the chairperson may simply have intended to refer to different authorities and that the legal principles expressed in the ruling remained sound.
However, the court rejected that argument.
Judge Allen-Yaman warned that if the chairperson had relied blindly on an AI assistant, such conduct would be unacceptable and contrary to standards already established by South African courts.
The judgement references several recent cases highlighting the dangers of AI-generated legal research.
It cites the 2023 Gauteng High Court decision in Parker v Forsyth, which cautioned that lawyers must apply an independent and questioning mind rather than “merely repeat in parrot-fashion, the unverified research of a chatbot”.
The court also drew on international precedent, quoting a recent English High Court ruling that warned AI tools can produce “apparently coherent and plausible responses” that may nevertheless contain fabricated sources or quotations.
It further cited a July 2026 judgement of the Supreme Court of India, which held that reliance on “non-existent, fake and hallucinated material” undermines the integrity of judicial decision-making and warrants setting aside the affected judgement.
Although the court stopped short of concluding that AI had definitely been used, it found there was sufficient evidence to question the chairperson's reasoning.
“The source of the erroneous citations has yet to be determined,” Judge Allen-Yaman said. “If they were indeed generated by an AI ‘assistant’, the ruling cannot stand.”
The judge added that even if the errors were made by the chairperson rather than AI, the applicants had afforded him an opportunity to explain the citations, which he failed to do.
The court concluded there was prima facie evidence that the chairperson had failed to properly apply his mind to the legal issues before him.
Judge Allen-Yaman said it would not serve the interests of justice to require the employees to continue appearing before a chairperson whose decision-making process had been called into serious question.
Continuing with the disciplinary hearing also risked wasting the time and resources of both parties should the rulings later be set aside.
Finding that exceptional circumstances existed, the Labour Court stayed the disciplinary proceedings pending the outcome of the review application. It made no order as to costs.
Share
