Authored By: Tsholofelo Makgale
I. Introduction
In recent years, generative artificial intelligence (AI) has rapidly entered legal practice, enabling legal practitioners to draft legal documents such as contracts, pleadings and company policies, conduct legal research, and analyse documents more efficiently. While these AI technologies promise greater efficiency and improved access to legal services, they also raise difficult questions concerning professional competence and responsibility. These benefits are accompanied by growing concern that excessive reliance on generative AI may compromise the competence expected of legal practitioners, particularly where hallucinated case law or fabricated legal authorities are presented to a court in support of legal arguments. The question then becomes: if AI generates incorrect legal authorities or flawed legal advice, where does accountability lie?
Although generative AI is transforming the delivery of legal services, its use does not and should not diminish the professional obligations imposed upon legal practitioners by South African law. Rather, it reinforces the continuing importance of professional competence, independent legal judgment, and accountability. Existing professional duties therefore remain applicable, but the increasing use of AI raises the question of whether the current legal framework provides sufficient guidance for its responsible use. This article accordingly evaluates the extent to which the existing South African legal framework governing professional competence adequately regulates the use of generative AI by legal practitioners.
Although the legal implications arising from the use of generative AI in legal practice are extensive, this article is confined to the ethical and legal responsibilities of legal practitioners insofar as professional competence is concerned.
II. The Existing Legal Framework Governing Professional Competence
A. Professional Competence under South African Law
The Legal Practice Act 28 of 2014 (LPA), read together with its rules, regulations, and code of conduct, constitutes the primary legal framework governing the legal profession in South Africa.1 Although enacted before the widespread adoption of generative AI in legal practice, the LPA establishes professional duties that remain applicable irrespective of the tools used by legal practitioners. The question is therefore whether these duties, particularly those relating to professional competence, are sufficiently clear to regulate AI-assisted legal practice. Within this framework, professional competence constitutes a core obligation requiring legal practitioners to render legal services with the knowledge, skill, diligence, and independent professional judgment expected of the profession.
The Code of Conduct requires legal practitioners, including candidate legal practitioners and juristic entities, to “maintain the highest standards of honesty and integrity.”2 In Council v Dladla, the court reaffirmed that the protection of the public remains the primary consideration, emphasising that legal practitioners must prioritise their clients’ interests while applying their best efforts in performing their duties competently and efficiently.3 These obligations are expressed in technology-neutral terms and therefore remain applicable irrespective of whether legal work is performed manually or with the assistance of generative AI. Accordingly, the conduct of legal practitioners remains subject to their duties to the court, the interests of justice, observance of the law, and the maintenance of appropriate ethical standards.4 In the absence of express statutory or professional guidance governing the use of generative AI by legal practitioners, these existing duties provide the primary normative framework within which AI-assisted legal practice must be assessed. Accordingly, practitioners remain responsible for ensuring that their conduct complies with the law and that any advice provided to clients or employers reflects their own independent and impartial professional judgment.5 Reliance on AI-generated outputs cannot displace this responsibility.
In addition to being required to demonstrate competence and efficiency in their work, legal practitioners are cautioned against undertaking work which they reasonably believe would render them incapable of carrying out in this manner using their best efforts.6 The obligation to employ one’s best efforts therefore requires that generative AI remain an assistive tool rather than a substitute for the legal practitioner’s own knowledge, judgment, and professional skill. Responsibility for the quality and accuracy of legal work therefore remains with the practitioner. Professional competence further requires legal practitioners to remain informed of legal developments, applicable legislation, legal theory, the common law, and developments within their respective fields of practice.7 Where generative AI is used in legal practice, this obligation arguably extends to understanding both its capabilities and its limitations so that inaccurate, incomplete, or fabricated outputs can be identified before they influence legal advice or court proceedings. Where the law has yet to address emerging technologies comprehensively, legal practitioners are expected to exercise heightened caution and refrain from conduct that could undermine the integrity, dignity, or reputation of the legal profession.8 The absence of AI-specific regulation does not diminish this responsibility; rather, it requires legal practitioners to exercise heightened professional caution when adopting emerging technologies.
In addition to strict observance of the code of conduct and its provisions, legal practitioners also have the responsibility to report unprofessional conduct by other legal practitioners, in order to commence disciplinary procedure.9 This responsibility assumes particular importance where practitioners become aware that AI-generated material has been relied upon without appropriate verification, thereby raising concerns regarding professional competence and ethical accountability.10
Consequently, a legal practitioner shall not mislead the court, whether directly or indirectly, on any matter of fact, question of law or content of the papers submitted to the court, including any transcript of evidence. Legal practitioners are strictly required to take all reasonable steps to avoid such action.11 Also, they may not rely on incorrect or false statements made in evidence and known to be so.12 I contend that the underlying responsibility or duty here is verification of the particular statements, rather than merely repeating what is reasonably believed to be true.
While the LPA and the Code of Conduct establish the legal framework governing professional competence, judicial interpretation has further defined the content and application of these obligations. An analysis of relevant case law therefore provides further insight into the standard of competence expected of legal practitioners.
B. Judicial Interpretation of Professional Conduct
Judicial decisions demonstrate that professional competence extends beyond technical legal knowledge to include honesty, diligence, independent judgment, professionalism, accountability, and the proper administration of justice. These principles provide a useful benchmark for assessing whether existing professional standards adequately regulate the use of generative AI in legal practice.
The first known case to address the usage of artificial intelligence in South African legal practice is Parker v Forsyth,13 where the plaintiff’s attorney had sourced the cases relied upon through ChatGPT. Through its use, the plaintiff’s attorneys conducted legal research which produced fictitious cases, entailing citations, facts and decisions.14 The issue of contention here, as it would seem, was not necessarily the use of this generative medium or tool, but the attorney’s conduct in merely accepting the generated results without verifying the accuracy thereof.15 The fabricated authorities were, however, detected before reaching the court, and not submitted as binding authority.16 The court did not view their proposed action as an attempt to mislead the court, but as simply being “overzealous and careless” in their omission to verify the generated legal research.17 This incident served as a timely reminder to lawyers not to rely on unverified research from generative AI as a substitute for independent legal research, but rather to combine the supposed efficiency of modern technology with independent reading.18 It was the court’s view that the embarrassment associated with the incident was probably sufficient punishment for the plaintiff’s attorneys.19 The significance of Parker v Forsyth, therefore, lies not in condemning the use of generative AI itself, but in reaffirming that legal practitioners remain personally responsible for verifying AI-generated research before relying upon it.
It was discovered in the case of Mavundla v MEC that all but one of the legal authorities referenced and relied upon were non-existent, incorrectly cited and taken out of context.20 This error was made by a candidate legal practitioner who denied having retrieved the references through generative AI, claiming instead to have obtained them in a legal journal which remains unnamed.21 The implication of this issue was not particularly premised on the use of generative AI, but on the conduct of legal practitioners, including candidate legal practitioners, and the duty they owed to the court which entails “honesty, candour and competence” and, further, not to mislead the court.22 The court referred extensively to Van Eck’s analysis, endorsing the view that although generative AI may improve efficiency, its limitations do not diminish legal practitioners’ existing ethical and professional responsibilities.23 The Parker v Forsyth case was also discussed at length in this judgment, with the court affirming that the attorney’s conduct was overlooked and that no directive was issued with regard “to the ethical and professional duties of legal practitioners in the use of ChatGPT.”24 Indeed, the court’s approach in Parker v Forsyth was said to “set a dangerous precedent”, given that the attorney’s careless conduct had not been considered as potentially amounting to negligence.25 The code of conduct of the legal profession remains the source from which the ethical and professional standards of legal practitioners can be extrapolated.26 Further, the rules’ silence on the limitations of AI technologies does not diminish legal practitioners’ ethical and professional duties.27 The breach in question in this case relates to rule 18.3 of the code of conduct, which requires that attorneys “exercise proper control and supervision” over their staff or offices. In this regard, the court affirmed that the attorney’s supervisory role over the candidate attorney as part of staff “would include the verification of the accuracy and correctness of any information sourced from generative AI systems.”28 This court deemed reliance on AI technologies for legal research to be irresponsible and unprofessional.29 The circumstances in this case were found to be more compelling than in Parker v Forsyth, since these fictitious and incorrect authorities were both issued and served; the court therefore ordered punitive costs on the irresponsible attorneys and referred the judgment to the LPC for further investigation and action.30
The case of Northbound Processing (Pty) Ltd v South African Diamond found the applicant’s attorneys to have relied on AI-hallucinated cases in their heads of argument, where the attorney admitted to neglecting to verify the accuracy of the citations.31 This case reinforces the approach adopted in Mavundla. The court again held that legal practitioners remain responsible for verifying AI-generated authorities and that negligent reliance on fabricated authorities constitutes a breach of professional obligations and their duty owed to the court.32 The court in this instance found the conduct of the applicant’s legal practitioners as constituting negligence, and made the same order as in Mavundla, namely referring such conduct to the LPC for investigation.33
The latest case in relation to the use of generative AI in legal practice is Roux v Greunen,34 which was delivered on 29 January 2026. The applicant in this instance is a lay litigant, not a legal practitioner. To bridge his lack of legal knowledge and understanding, the applicant admitted to using various AI systems for research and legal analysis to make his legal submissions.35 Unfortunately, these papers contained “irrelevant matter, ineffective legal analysis, and contentions based on unclear principles not established in South African law, as well as AI hallucinated case law.”36 The court acknowledged that artificial intelligence can be a useful tool when used properly; the same cannot be said, however, when it is used by someone with insufficient knowledge and understanding of the law and its processes.37 In this regard, both legal practitioners and lay litigants ought to exercise caution and verify AI-generated information and advice before acting on it.38 The main application was dismissed with punitive costs ordered against the applicant.39
Although Council v Dladla is not related to generative AI, it reinforces the underlying principle that where a legal practitioner’s conduct falls below the required standard of competence, both the courts and the LPC have a duty to intervene in order to preserve the integrity of the courts and legal profession.40 Of particular importance, however, is the preservation of public confidence in the administration of justice. The cases discussed above demonstrate the application of this established principle to emerging technologies, rather than the creation of an entirely new professional standard.
Collectively, these decisions demonstrate that South African courts have not regarded the use of generative AI as inherently improper or unlawful. Rather, they have consistently held that existing duties of professional competence, honesty, independent judgment, supervision, and accountability remain fully applicable where AI is used. This judicial interpretation, therefore, suggests that the current legal framework is capable of regulating AI-assisted legal practice through technology-neutral professional standards. At the same time, the differing factual circumstances and judicial responses indicate that further guidance may be necessary to assist legal practitioners in applying these duties consistently when using generative AI.
III. Generative Artificial Intelligence and Professional Competence
A. The Capabilities and Limitations of Generative Artificial Intelligence
The use of generative AI in legal practice is increasing worldwide due to its efficiency in legal research, document drafting, and client services.41 The unreported Makunga v Barlequins Bellegings case illustrates AI’s potential to promote access to justice in assisting self-represented litigants with enhancing legal arguments.42 While generative AI is capable of handling routine tasks, such as summarising legal texts and refining language, amongst others, it remains problematic when used in complex legal reasoning or applying law to specific facts.43 Another scholar similarly argues that this simply means that AI has the potential to assist human work rather than replace it, because while technology has a transformative influence on the work of lawyers, it remains incomplete.44 Therefore, its area of fundamental efficiency and effectiveness lies in sifting through vast amounts of data, and routine, repeated and administrative tasks.45
In stark contrast, AI-powered tools in legal practice all create risks related to misinformation, bias, confidentiality breaches, and the erosion of professional judgment.46 The South African courts have illustrated those risks.47 The principal risks and key limitations of generative AI models are their unreliable retrieval and application of the law, as well as misrepresentations and fabricated citations.48 AI is, therefore, best used as an assistant and not an authoritative source.49 Despite the self-learning abilities of AI technologies, human verification remains indispensable.50 The literature consistently emphasises that AI-generated outputs should remain subject to meaningful human oversight and verification, particularly where they inform legal advice, legal research, or court submissions.
AI models are mainly trained and developed on European legal traditions and frameworks.51 This raises the question of whether AI is able to engage meaningfully with South African law.52 Because most AI models are trained predominantly on foreign legal materials, their application to South African law remains limited and requires careful verification.53 This increases the risk of inaccurate citations and the inappropriate application of foreign legal principles.54
Collectively, the literature demonstrates that the value of generative AI lies in its ability to enhance efficiency rather than replace legal expertise. Its limitations reinforce the continuing necessity for legal practitioners to exercise independent judgment, verify AI-generated outputs, and remain accountable for the legal services they provide. These findings support the continued relevance of existing professional competence obligations in the context of AI-assisted legal practice.
IV. Does the Existing Legal Framework Adequately Regulate Artificial Intelligence
A. Is reform necessary?
Although the existing legal framework governing professional competence is sufficiently broad to regulate the use of generative AI, the absence of AI-specific professional guidance creates uncertainty regarding the practical application of these duties. The question is therefore not whether entirely new professional obligations are required, but whether clearer guidance is necessary to assist legal practitioners in complying with their existing duties.
Increased reliance by legal practitioners on the use of AI may lead to difficulties in identifying and correcting errors.55 This necessitates the constant application of legal skills to AI-generated content, in order to develop and maintain those skills.56 Several scholars have proposed practical ethical guidelines that reinforce, rather than replace, existing professional duties. These include requiring legal practitioners to assume responsibility for AI-assisted work, exercise independent judgment when using AI for substantive legal analysis, preserve confidentiality, and ensure that AI enhances rather than compromises client interests.57
Given that many generative AI systems are developed and trained primarily on foreign legal materials, any future guidance issued by the LPC should require legal practitioners to critically evaluate AI-generated outputs for bias, jurisdictional accuracy, and consistency with South African constitutional values and legal principles before relying on them.
V. Conclusion
The existing South African legal framework governing professional competence is largely capable of regulating the use of generative AI, as its technology-neutral duties continue to apply irrespective of the tools employed by legal practitioners. Recent judicial decisions demonstrate that professional competence, honesty, independent judgment, and accountability remain the governing standards where AI is used. Legal practitioners therefore remain responsible for verifying AI-generated outputs and ensuring that the use of generative AI enhances, rather than compromises, the administration of justice. Nevertheless, the absence of profession-specific guidance creates practical uncertainty. Rather than requiring fundamental legislative reform, the responsible use of generative AI would be better supported through AI-specific guidance issued by the Legal Practice Council, together with continued judicial development and practitioner education.
Note(S):
1. Legal Practice Act 28 of 2014.
2. Legal Practice Council, Code of Conduct for Legal Practitioners, Candidate Legal Practitioners and Juristic Entities 2019, para 3.1.
3. Council v Dladla (5849/21) [2022] ZAGPPHC 920 (22 November 2022) [23].
4. Legal Practice Council, Code of Conduct, para 3.3.
5. Legal Practice Council, Code of Conduct, paras 3.5, 3.9.
6. Legal Practice Council, Code of Conduct, para 3.11.
7. Legal Practice Council, Code of Conduct, para 3.13.
8. Legal Practice Council, Code of Conduct, para 3.15.
9. Legal Practice Council, Code of Conduct, paras 4, 5.
10. Legal Practice Council, Code of Conduct, para 57.1.
11. Legal Practice Council, Code of Conduct, para 57.1.
12. Legal Practice Council, Code of Conduct, para 57.9.
13. Parker v Forsyth NNO and Others (1585/20) [2023] ZAGPRD 1 (29 June 2023).
14. Ibid [87].
15. Ibid [87].
16. Ibid [89].
17. Ibid [89].
18. Ibid [90].
19. Ibid [91].
20. Mavundla v MEC: Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal and Others (7940/2024P) [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP) (8 January 2025) [41].
21. Ibid [21], [40].
22. Ibid [41].
23. Ibid [44]-[49].
24. Ibid [44].
25. Ibid.
26. Ibid [45]-[46].
27. Ibid [46]-[47].
28. Ibid [48].
29. Ibid [50].
30. Ibid [51]-[55].
31. Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator and Others (2025/072038) [2025] ZAGPJHC 661 (30 June 2025) [89]-[91].
32. Ibid [92]-[93].
33. Ibid [95]-[96].
34. Roux v Greunen and Others (2025/81170) [2026] ZAGPJHC 55 (29 January 2026).
35. Ibid [119].
36. Ibid [119]-[126].
37. Ibid [120].
38. Ibid.
39. Ibid [159].
40. Council v Dladla (n 3) [25].
41. Donrich Thaldar and others, “Responsible AI use in South African legal practice: A call for ethical guidelines” [2025] (663) De Rebus 18, 18.
42. Ibid.
43. Ibid.
44. Willem H Gravett “Is the Dawn of the Robot Lawyer upon us? The Fourth Industrial Revolution and the Future of Lawyers” (2020) (23) PER / PELJ 2020 1, 4.
45. Ibid 16.
46. Thaldar and others (n 41) 18, 18.
47. Daniel Van der Merwe “A Local, Continental (African) and International Overview of the Law as it Relates (or Tries to Relate) to Artificial Intelligence (AI)” 2024 (27) PER / PELJ 1, 21.
48. Donrich Thaldar “Can AI Think Like a Lawyer? Evaluating Generative AI in South African Law” (2025) (28) PER / PELJ 1, 1.
49. Thaldar and others (n 41) 18, 18.
50. Van der Merwe (n 47) 1, 13.
51. Thaldar and others (n 41) 18, 19.
52. Thaldar (n 48) 1, 2.
53. Van der Merwe (n 47) 1, 21.
54. Thaldar and others (n 41) 18, 19.
55. Gravett (n 44) 1, 26.
56. Ibid.
57. Thaldar and others (n 41) 18, 18-19.
Bibliography
Table of Cases
Council v Dladla (5849/21) [2022] ZAGPPHC 920 (22 November 2022)
Mavundla v MEC: Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal and Others (7940/2024P) [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP) (8 January 2025)
Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator and Others (2025/072038) [2025] ZAGPJHC 661 (30 June 2025)
Roux v Greunen and Others (2025/81170) [2026] ZAGPJHC 55 (29 January 2026)
Parker v Forsyth NNO and Others (1585/20) [2023] ZAGPRD 1 (29 June 2023)
Table of Legislation
Legal Practice Act 28 of 2014
Legal Practice Council, Code of Conduct for All Legal Practitioners, Candidate Legal Practitioners and Juristic Entities 2019
Bibliography
Gravett WH “Is the Dawn of the Robot Lawyer upon us? The Fourth Industrial Revolution and the Future of Lawyers” (2020) (23) PER / PELJ 2020 1
Thaldar D “Can AI Think Like a Lawyer? Evaluating Generative AI in South African Law” (2025) (28) PER / PELJ 1
Thaldar D and others, “Responsible AI use in South African legal practice: A call for ethical guidelines” (2025) (663) De Rebus 18
Van der Merwe D “A Local, Continental (African) and International Overview of the Law as it Relates (or Tries to Relate) to Artificial Intelligence (AI)” (2024) (27) PER / PELJ 1
