South African Courts Grapple With AI Use Amid Unfinalized Policy

AI is increasingly finding its way into South Africa’s courts. (Image created with GenAI) Amid the increasing use of artificial intelligence (AI) by judges to aid them in making determinations – a move that follows lawyers presenting hallucinated cases – a draft policy dealing with the use of this technology has yet to be finalised. During Judicial Service Commission interviews on Monday, KwaZulu-Natal High Court Judge Mokgere Masipa said the draft policy had been presented at this year’s Judges’ Conference and was being finalised after concerns were raised around governance and data security, EWN reported . The conference , held in July, resolved that the Judiciary Administration and Information Technology Steering Committee should finalise the Draft Policy on Artificial Intelligence. It also called for clear rules on permissible and prohibited uses of AI, with human oversight and judicial decision-making remaining central. Alistair Pugin, founder of cloud technology enabler Celyntis, says AI is already being adopted across the legal sector, but its use is moving faster than the development of formal standards. “It’s not in the future; it’s here right now,” Pugin says. Governance is critical and “there has to be a framework for using AI in the legal system. There have to be standards defined for this as well.” World Wide Worx MD Arthur Goldstuck says he sees “real value in using AI to reconcile evidence, provided the judge checks the work against the original records. That includes checking what the AI left out and any assumptions it made along the way.” Yet, it is too early to conclude that South African judges are broadly embracing generative AI, says Vanessa Manson, consultant and admitted attorney at law firm Ellipsis. She points to FJL v TGO , in which Judge Ingrid Opperman drew a distinction between AI assisting judges and being used to decide a matter. “The key distinction… is between AI as an assistant and AI as a decision-maker or verifier of last instance,” says Opperman. “Responsibility for accuracy in all respects rests with the author,” the judge adds. Various court cases have seen judges experimenting with AI. (Graphic: Nicola Mawson) Manson adds: “AI may be used to help with tasks such as summarising information, conducting research, editing, administrative work and drafting, provided that the judge checks the AI’s work and remains fully responsible for the judgement.” The judiciary, in a December 2025 newsletter , flagged risks around judges using public AI platforms, such as ChatGPT, including privacy concerns and the need for judges to remain in control of their rulings. In Mavundla v MEC , Judge E Bezuidenhout turned to ChatGPT after discovering that case law cited in an application for leave to appeal against an earlier judgement could not be verified. Of nine cases cited in the application, only two existed, and one of those had been incorrectly cited. Other authorities either did not exist, or did not support the arguments for which they had been used, the January 2025 case found. The judge notes that the candidate legal practitioner who conducted the research denied using ChatGPT. The Judiciary Administration and Information Technology Steering Committee has yet to finalise the Draft Policy on Artificial Intelligence. (Graphic: Nicola Mawson) Bezuidenhout then decided to test the technology. “By way of a brief experiment, the ChatGPT application was used and the citation of the Pieterse [case, among those relied upon by the applicants] was entered,” the judge said. ChatGPT verified that the case existed and said it was relevant to the issue for which it had been cited; claims that turned out to be false. This, Bezuidenhout said, “immediately illustrated the unreliability of it as a source of information and legal research”. The bot also confirmed the existence of another case, including the name of the supposed judge, that the court could not find. “In my view, relying on AI technologies when doing lega