South Africa: Judicial AI Policy Draft Nears Finalization

• South African courts are increasingly seeing AI use by judges and lawyers, prompting the development of a national policy.
• The Draft Policy on Artificial Intelligence for the judiciary is being finalized by the Judiciary Administration and Information Technology Steering Committee following concerns about governance and data security.
• Judges have highlighted AI's unreliability, with cases like `Mavundla v MEC` demonstrating "hallucinations" and false legal citations from platforms like ChatGPT.
• Judicial guidance emphasizes AI as an assistant tool for tasks like research and drafting, but stresses that human judges retain full responsibility for accuracy and decision-making.
• Experts underscore the urgent need for formal standards and frameworks to govern AI use in the legal sector, as its adoption outpaces regulation.

The experiment immediately demonstrated the unreliability of AI as a source for information and legal research, highlighting the critical need for human verification.

Artificial intelligence is increasingly integrated into South Africa's judicial landscape, with judges employing the technology to assist in their determinations. This growing adoption comes amidst instances where legal practitioners have presented cases generated by AI that contained fabricated information. In response to this evolving environment, a `South Africa judicial AI policy draft` is currently undergoing finalization to establish clear guidelines for the technology's use.

During recent Judicial Service Commission interviews, KwaZulu-Natal High Court Judge Mokgere Masipa confirmed that this draft policy had been presented at the Judges' Conference held in July of this year. The policy's finalization is being expedited due to concerns that have been raised regarding governance and data security within the judicial system. Experts in cloud technology, such as Alistair Pugin, founder of Celyntis, emphasize that AI is already a present reality across the legal sector, with its implementation outpacing the development of formal regulatory standards. Pugin stresses the critical need for robust governance and a defined framework to guide AI's application in the legal system, asserting that specific standards must be established.

The Judges' Conference, convened in July, specifically resolved that the Judiciary Administration and Information Technology Steering Committee should take responsibility for finalizing the `Draft Policy on Artificial Intelligence judiciary`. This resolution underscored the necessity for explicit rules delineating both permissible and prohibited uses of AI within court proceedings. A core principle guiding this policy development is the unwavering centrality of human oversight and judicial decision-making, ensuring that technology remains a tool rather than a substitute for human judgment.

This proactive approach to `South Africa legal AI governance` aims to address the complexities introduced by AI. The policy seeks to provide clarity for all stakeholders, from judges to legal practitioners, on how to ethically and effectively integrate AI tools while safeguarding the integrity of the judicial process. The ongoing work by the Judiciary Administration and Information Technology Steering Committee is crucial for establishing these foundational guidelines, which will shape the future interaction between AI and the South African legal system.

Despite the increasing presence of AI, it is premature to conclude that South African judges are broadly embracing generative AI, according to Vanessa Manson, a consultant and attorney at Ellipsis. Judicial pronouncements have highlighted significant distinctions and risks associated with `South Africa court AI use`. In the case of `FJL v TGO AI judgment`, Judge Ingrid Opperman drew a clear line between AI serving as an assistant to judges and AI being used to make or verify decisions. Judge Opperman unequivocally stated that the responsibility for accuracy in all aspects rests solely with the author.

Further illustrating the potential pitfalls, a December 2025 judiciary newsletter cautioned against judges using public AI platforms like ChatGPT, citing privacy concerns and the imperative for judges to maintain full control over their rulings. The `Mavundla v MEC ChatGPT reliability` case, heard in January 2025, provided a stark example. Judge E Bezuidenhout turned to ChatGPT after discovering that nine case law citations in an application for leave to appeal were largely unverified or non-existent. Only two cases were found to exist, one of which was incorrectly cited, and other authorities either did not exist or failed to support the arguments presented. The candidate legal practitioner involved denied using ChatGPT for the research.

In an experiment, Judge Bezuidenhout entered a citation from the unverified cases into ChatGPT. The AI falsely confirmed the case's existence and relevance. The bot also confirmed another non-existent case, even providing a fictitious judge's name. The experiment immediately demonstrated the unreliability of AI as a source for information and legal research, highlighting the critical need for human verification.

Legal technology experts underscore the immediate need for robust frameworks governing AI in the legal sector. Alistair Pugin reiterates that AI's adoption is moving faster than the development of formal standards, making governance critical. Arthur Goldstuck, MD of World Wide Worx, acknowledges the significant value AI can offer in reconciling evidence, provided that judges meticulously verify the AI's output against original records, paying close attention to any omissions or assumptions made by the technology.

Vanessa Manson further clarifies that AI's permissible uses in the judiciary generally extend to tasks such as summarising information, conducting research, editing, administrative work, and drafting. However, she stresses that in all such applications, the judge must thoroughly check the AI's work and retain full responsibility for the final judgment. This collective expert sentiment reinforces the cautious yet pragmatic approach being taken towards the `Judicial Service Commission AI policy`, emphasizing that while AI can be a powerful tool, it must always operate under stringent human oversight and accountability.

Lawyers and compliance officers should note the ongoing development of the Draft Policy on Artificial Intelligence for the South African judiciary, particularly given the documented risks of AI 'hallucinations' and unverified citations in court. This necessitates rigorous verification of any AI-generated legal research and careful monitoring of the policy's finalisation to understand future permissible and prohibited uses of AI in court proceedings.

Source: Original reporting via EWN