11 September 2026 7 min read
Will AI replace attorneys? What South African courts are already saying
Every few months a new headline asks whether artificial intelligence will make lawyers obsolete. It is a fair question. Tools that draft, summarise and research in seconds are now cheap and everywhere, and a profession built on reading and writing is an obvious target.
South African courts have started giving a practical answer. Three judgments in three years show where the line sits, and it is not where the headlines put it.
Three cases, one lesson
Parker v Forsyth NNO (2023). In a defamation matter before a regional court in Johannesburg, the plaintiff's legal team relied on case authorities that turned out not to exist. They had been produced by ChatGPT. The court accepted that there was no intention to mislead, but it was plainly unimpressed: the team had relied on a chatbot's research without checking it, wasting the court's time and the other side's. Commentators report a punitive costs order for the period in which the fictitious cases were persisted with.
Mavundla v MEC: Department of Co-operative Government and Traditional Affairs, KwaZulu-Natal (2025). In an application for leave to appeal, seven of the nine cases the applicant's team relied on did not exist, and the other two were cited incorrectly. The team did not verify its references even when given the chance. The judge found it highly likely that generative AI had been used, dismissed the application and referred the matter to the Legal Practice Council.
Northbound Processing v South African Diamond and Precious Metals Regulator (2025). In an urgent application in the Gauteng High Court, two cases cited in counsel's heads of argument were found to be fictitious while the judgment was being written. Counsel eventually explained that they had used an online subscription tool marketed as trained on South African law. Both senior and junior counsel apologised, and the court accepted that oral argument had not relied on the false citations. It granted the relief sought, and still referred the legal team to the Legal Practice Council.
What the courts did not say
None of these judgments says lawyers may not use AI. Not one of them treats the technology itself as the problem.
What they say is narrower and more useful: the lawyer answers for every word filed in their name. It does not matter whether a draft came from a candidate attorney, a precedent bank or a language model. If it reaches the court, the practitioner is responsible for it being true. Northbound is the sharpest example. The tool was a paid product built for South African law, used by experienced counsel, and the result was still a referral.
The same principle runs through draft ethics guidelines on generative AI prepared by academics at the University of KwaZulu-Natal for submission to the Law Society of South Africa. As reported, the draft's starting point is that practitioners remain fully responsible for work produced with AI's help, and that AI output should be checked, never simply trusted.
So will AI replace attorneys?
The cases suggest a different question: which parts of an attorney's work can be handed to software, and which cannot?
Judgement cannot. Deciding what a client should do, weighing risk, reading a judge or an opposing party: these depend on experience and accountability that a model does not have and cannot carry.
Advocacy cannot. Standing up in court, negotiating, persuading a client to take advice they do not want to hear.
Responsibility cannot. This is the point the courts keep making. A model can produce a draft. It cannot be struck off.
What software can do, and increasingly does well, is the groundwork around that core: first drafts, summaries of long documents, transcription, finding the right precedent to start from, and the administrative work that surrounds every matter. Used properly, with a lawyer checking the output, that is not a threat to the profession. It is time handed back to it.
The real change is to the shape of the day
Thomson Reuters' Future of Professionals report for 2025, based on a survey of 2,275 professionals in law, tax, accounting and related fields, found that respondents expect AI to free up about five hours a week within the next year. That is a forecast from the people doing the work, not a measured result, but it points in a clear direction.
For a firm that bills by the hour, five hours a week is not a small number. Some of it will go into better work. Some will go into the matters and clients that never get enough attention. Some of it, ideally, will go home.
The firms that benefit most will be the ones that are deliberate about it: clear about which tasks AI may help with, strict about verification, and honest with clients about how their work is done.
Where we stand
We build Atora on the same principle the courts keep returning to. Atora drafts time entries from the work a lawyer has done, and the lawyer approves every one of them. Nothing reaches a client's bill without a person deciding it should. AI does the remembering and the typing. The judgement stays with you.
AI will not replace attorneys. It will, however, change what a good day's work looks like. The question for every firm is whether that change happens to them or for them.
This article is general information, not legal advice.
Sources
- Cliffe Dekker Hofmeyr, Fictional citations, real consequences: a cautionary tale for the modern lawyer (January 2025), on Mavundla [2025] ZAKZPHC 2
- Cliffe Dekker Hofmeyr, South African courts show no tolerance for AI hallucinated cases (July 2025), on Northbound Processing
- Michalsons, Northbound Processing v SA Diamond Regulator: AI-generated case law
- Michalsons, Parker v Forsyth: lessons for using AI for legal research
- University of KwaZulu-Natal, Law academics draft AI ethics guidelines for legal practitioners
- Thomson Reuters, Future of Professionals Report 2025