A South African junior advocate who relied on an artificial intelligence tool to draft court submissions has been referred to the Legal Practice Council for investigation after including fictitious case law in written arguments. The case, involving a disputed licence linked to the sale of Rappa Resources to Northbound Processing, has become one of the country’s most prominent legal episodes involving the misuse of generative AI.
Northbound Processing had approached the court on an urgent basis, asking the South African Diamond and Precious Metals Regulator to release a refining licence, arguing it faced severe commercial losses without it. While the court ultimately ruled in Northbound’s favour, Acting Judge DJ Smit referred the matter to legal regulators after discovering multiple non-existent case citations in the heads of argument submitted by Northbound’s legal team.
“In my view, it matters not that such cases were not presented orally, but were contained in written heads of argument,” Smit wrote in his ruling. “Written heads are as important a memorial of counsel’s argument as oral argument and, for purely practical reasons, are often more heavily relied upon by judges.”
Junior Counsel Admits to Using AI Tool
When questioned, the junior counsel admitted to using Legal Genius, an AI tool marketed as being trained on South African case law, and explained that the errors stemmed from time constraints and the unavailability of the colleague who initially drafted the document. He apologised unreservedly and took full responsibility, while insisting there was no intent to mislead the court. Northbound’s senior counsel, Arnold Subel, also issued an apology, saying he had conducted only a “sense-check” on the document and believed the propositions were trite.
Screenshot of Legal Genius Website
Broader Warnings About AI in Legal Practice
Refilwe Motsoeneng, associate at digital and technology law firm Michalsons Giles Inc, noted that the case illustrates the legal and ethical risks of unverified AI-generated content in court.
“This case reinforces the ethical duty to independently verify all legal sources before citing them. Even unintentional AI ‘hallucinations’ can cause reputational harm and lead to professional misconduct complaints,” she said.
“They cannot replace proper legal research or professional oversight. Courts expect high standards, especially in urgent applications where they rely on written submissions,” she added.
Part of a Growing Pattern in South African Courts
This is now the third known instance in South Africa where large language models were found to have introduced false legal citations in court filings:
- In 2023, legal representatives for a Sectional Title Scheme cited fictitious authority from ChatGPT and received a punitive cost order.
- In 2025, Judge Elsje-Marie Bezuidenhout in a separate case found only two of nine citations in an appeal by Umvoti Mayor Godfrey Mavundla to be legitimate. She called the incident “irresponsible and downright unprofessional” and referred it to the Legal Practice Council.
Judge Smit echoed that ruling, referencing Article 16(1) of the Code of Judicial Conduct, which obliges judges to report “serious professional misconduct or gross incompetence” to relevant authorities.
Clarifying the Legal Principle in Focus
Beyond the AI controversy, the court’s judgment also confirmed a key regulatory point: once a licence is issued and communicated by a regulator, it cannot be unilaterally withheld without following proper legal processes.
“Unless a decision is formally challenged in court and set aside, it remains valid and must be acted on. This is important for any business that depends on government licences or approvals to operate,” said Motsoeneng.
The case is likely to serve as a cautionary precedent for legal professionals experimenting with generative AI in litigation, reinforcing the need for rigorous source verification and clear professional boundaries in digital legal practice.
