While concurring with the main ruling in FJL vs TGO (2025/220239) [2026] ZAGPJHC 875, Judge Opperman delivered a separate judgment dealing, in part, with how AI could be used and applied within the judiciary process. Drawing on research into international approaches, she emphasised the importance of the judiciary retaining governance over the development, procurement, and use of AI systems, so that human oversight and independent judicial reasoning remain central to their deployment.
Why Did AI Make it into the Judgment?
The issue arose after 11 discrepancies in an earlier High Court judgment were pointed out in one of the rulings in a long-running matter between a divorced couple concerning their children’s private schooling.
The discrepancies included an incorrect citation of a case, a citation to a case that does not exist, and references to judgments where, although the citation itself exists, the quotation did not deal with the matter at hand.
Both parties agreed that the respondent’s legal counsel had not provided the High Court with the incorrect references, despite the earlier judgment recording, in some instances, that she had.
The separate judgment reads: “The errors can therefore not be attributed to counsel. The most plausible explanation, certainly for the fictitious Lubbe reference, is that it is the product of the use of Artificial Intelligence (AI) and what has been dubbed ‘hallucinations’.”
Although the judge made no finding on whether AI was used in the earlier judgment, she stated: “Nevertheless, it was suggested in these proceedings and I am compelled to explain the implication of the use of AI and its impact on independent judicial reasoning, or, as it has been termed ‘the erosion of independent judicial reasoning’.”
Here she referenced a recent De Rebus article titled “Artificial intelligence in the South African judiciary: Friend or threat?”, authored by Lizelle Cloete LLB (UWC), a District Court Magistrate in Carnarvon.
Looking to International Examples
South Africa does not yet have its own AI guidelines for the judiciary, and one cannot say whether other courts will in future follow the approach set out in this separate concurring judgment. It does however provide a starting point as it references examples from jurisdictions such as the United Kingdom, United States, Canada and Australia.
According to the South African judge, she found no instances in any of these countries where AI as a decision-maker was acknowledged or countenanced, or where a judge has used AI to reason or determine the outcome of a judicial decision.
On the contrary, judicial guidance strongly suggests that such conduct would be unacceptable because judicial power is vested in human judges, they have a non-delegable duty to reason and decide cases and litigants are entitled to human adjudication.
AI Can be Used as an Assistant
The judge noted that all four jurisdictions make a distinction between AI as an assistant and AI as a decision-maker or “verifier of last instance”. It can assist, among others, with administrative efficiency, summarising, research, editing and drafting support.
She writes that these uses appear to be generally acceptable on the condition that the judge verifies the output and remains responsible for the final judgment.
‘Jolly Useful’ as Summarisation Tool
The separate concurring judgment mentions a UK example where a Lord Justice in the Court of Appeal publicly disclosed that he had used ChatGPT for a summary of an area of law while drafting a judgment. He described it as “jolly useful” with summarisation but was clear it was not used for decision-making.
The UK’s recent formal guidance on judicial use of AI similarly emphasises that judges remain personally responsible for all material issued in their name, that AI-generated material may contain inaccuracies, and that confidential information should not be entered into public AI systems. The guidance permits “carefully supervised use of AI but rejects any suggestion that judicial responsibility can be delegated to AI” and any use must preserve the integrity of the administration of justice and the rule of law.
Turning to Sanctions
The High Court judge recorded that, had she been able to conclude that the reasoning in the earlier judgment was AI-generated and the true decision-maker was a machine, she would have concluded that the appellant’s rights under section 34 of the Constitution had been infringed.
“To divest the decision-making responsibility to a robot, breaches not only the oath of office, which Acting Judges take, but it also denies a party the right to be heard by a Court as intended in terms of section 34.” (Emphasis in the original separate concurring judgment.)
According to the judge no reported American case has been identified where a judge delegated legal reasoning or the determination of a judgment’s outcome to generative AI. Numerous courts in the US have imposed sanctions, costs orders, and disciplinary referrals in cases where lawyers submitted fabricated AI-generated authorities.
Law360 Canada, in an article in June 2026, stated that when a Canadian court or tribunal finds that a party has relied on a case that does not exist, the consequences differed. In some cases, lawyers responsible have been ordered to pay costs personally, ranging from relatively modest to substantial amounts, while in others no sanctions have been imposed.
What Could This Mean for the Future Use of AI in the Judiciary?
From the international examples considered in the judgment it is evident that the use of AI in the judicial field and in courts will in future be an important area of developing law.
The separate concurring judgment could prove useful in this regard as it refers to a decision at the 2026 South African Judiciary Conference that the Draft Policy on Artificial Intelligenceshould be clear about the distinction between administrative and adjudicative functions and the permissible and prohibited uses of AI.
Further to this, the judiciary would leverage AI through customised and secure platforms, initially focusing on efficiency tools, including court recording and transcription, translation, case management, and document summarisation, to reduce delays and enhance access to justice.
The judiciary also committed to providing ongoing education on the responsible use of AI, while ensuring the continued development and maintenance of the critical legal research, analytical, reasoning and judgment-writing skills that underpin judicial decision-making.
Broader Questions for Professional Services
Following the judgment, industry experts have highlighted several issues that may warrant further consideration as professional services firms and practitioners increasingly use AI.
These centre around:
- Professionals, including legal professionals, who work on hourly rates: They probably need to address AI assistance in their engagement letters. Where AI materially reduces the time required to perform a task, they cannot charge for the actual hours that would previously have been worked, but in fact less hours were put in by using AI.
- The methodology and approach in which legal documents are “taxed” in the case of cost orders: This may require a rethink. There is no parity in charging a party for cost orders on a particular scale when the actual work was performed using the efficiency of AI tools.
- Professional indemnity and disclosure: Where professionals use AI, items such as professional indemnity will have to be updated, especially where the wrong results are achieved. Examples might be that the use of any AI should be disclosed in work produced, and that this will be excluded from insurance policies.
The separate concurring judgment could provide a useful starting point for considering how efficiency, accountability, costs and professional responsibility should be addressed alongside the opportunities and benefits presented by AI.
