Authored By: Nkagiseng Mpafudi
University of Witwatersrand
Introduction
Artificial intelligence (AI) is rapidly transforming the legal profession. Once viewed as a futuristic concept, AI is now routinely used by legal practitioners to conduct legal research, summarise judicial decisions, draft legal documents and analyse large volumes of information within a fraction of the time required by traditional methods. These technological advancements have undoubtedly improved efficiency and accessibility within legal practice. However, they have also introduced significant ethical and professional challenges, particularly where AI-generated information is inaccurate or entirely fabricated. The emergence of generative AI systems such as ChatGPT has therefore sparked an important legal question: who should bear responsibility when AI-generated errors influence legal proceedings?
This question is no longer hypothetical. Courts in several jurisdictions have already encountered cases in which legal practitioners relied upon fictitious authorities produced by AI. The widely publicised decision in Mata v Avianca, Inc.demonstrated the serious consequences of submitting AI-generated case law without proper verification, while the South African High Court reached a similar conclusion in Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others. These decisions illustrate that although artificial intelligence may assist legal practice, it cannot replace the professional judgment and ethical obligations expected of legal practitioners.[1]
Although some commentators argue that increasingly sophisticated AI systems may eventually require new models of legal liability, the current legal framework places responsibility on the legal practitioner who chooses to rely on AI-generated outputs.
This article argues that artificial intelligence cannot bear legal or professional responsibility because it lacks legal personality, independent legal duties and the capacity to exercise ethical judgment. Responsibility for AI-generated errors must therefore remain with legal practitioners, whose obligations of competence, honesty and diligence cannot be delegated to technology. It is submitted that while AI should be embraced as an innovative legal tool capable of improving access to justice and enhancing legal efficiency, its use must be regulated through clear ethical standards that preserve accountability, protect the integrity of judicial proceedings and uphold the rule of law.
Artificial Intelligence as an Assistive Legal Tool
Artificial intelligence has transformed the delivery of legal services by enabling practitioners to perform tasks that were previously time-consuming and labour-intensive. AI-powered systems can analyse contracts, identify relevant precedents, organise documentary evidence and assist with legal drafting within seconds. Richard Susskind argues that technological innovation will continue to reshape legal services by improving efficiency and expanding access to legal assistance, particularly where routine legal tasks can be automated. These developments have encouraged many legal practitioners to incorporate AI into their daily professional activities.[2]
Unlike traditional legal research databases such as LexisNexis and Westlaw, generative artificial intelligence does not independently verify the legal accuracy of its outputs and may produce fictitious authorities, commonly referred to as “AI hallucinations”. Consequently, practitioners must independently verify AI-generated information before relying upon it in legal proceedings.[3]
These risks were illustrated in Mata v Avianca, Inc., where attorneys relied on fictitious authorities generated by ChatGPT. The United States District Court held that lawyers remain personally responsible for verifying every authority presented before the court. The decision reaffirmed that AI may assist legal research but cannot replace professional judgment or ethical responsibility.[4]
Professional Responsibility Under South African Law
The principle established in Mata has been reinforced within South African jurisprudence. In Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others, the KwaZulu-Natal High Court was confronted with legal submissions containing authorities generated through ChatGPT that had not been properly verified. The court strongly criticised this conduct, describing reliance on unverified AI-generated authorities as “irresponsible and downright unprofessional”.[5] More importantly, the court emphasised that ignorance of the limitations of artificial intelligence does not excuse a failure to comply with the ethical obligations imposed upon legal practitioners. The judgment makes it clear that while AI may be used as a research aid, it cannot replace the professional responsibility to verify legal authorities before they are presented to a court.
Professional responsibility presupposes legal personality and enforceable ethical duties. Because AI possesses neither, it cannot be admitted as a legal practitioner, owe duties to clients or be disciplined by the Legal Practice Council. Responsibility for AI-generated legal work must therefore remain with the practitioner who relies upon AI.
This reasoning is consistent with the earlier decision in Parker v Forsyth NO and Others, where legal representatives similarly relied upon fictitious authorities generated by ChatGPT. Although Parker is an unreported decision, it was expressly discussed and endorsed by the High Court in Mavundla, demonstrating the growing judicial concern regarding the misuse of generative AI in legal proceedings.[6] Together, these decisions establish an emerging principle within South African law: technological innovation cannot diminish professional accountability.
The South African legislative framework further supports this conclusion. The Legal Practice Act 28 of 2014 seeks to promote an independent, ethical and accountable legal profession that serves the public interest.[7] The Act is complemented by the Code of Conduct for all Legal Practitioners, Candidate Legal Practitioners and Juristic Entities, which requires practitioners to maintain the highest standards of honesty, integrity, competence and diligence while avoiding conduct that misleads the court.[8] These duties are not diminished merely because legal work is produced with the assistance of artificial intelligence. If a legal practitioner files documents containing fabricated authorities or inaccurate legal propositions generated by AI, responsibility rests with the practitioner who chose to rely upon that information without adequate verification.
The constitutional framework reinforces this position. Section 1(c) of the Constitution establishes the supremacy of the Constitution and the rule of law as foundational values of South Africa’s constitutional democracy.[9] The rule of law requires judicial decisions to be based upon authentic legal authorities and reliable legal reasoning rather than fabricated or unverifiable information. Similarly, section 34 guarantees every person the right to have legal disputes resolved in a fair public hearing before an independent and impartial court, while section 165 protects the independence, dignity and effectiveness of the judiciary.[10] These constitutional guarantees would be undermined if courts were expected to rely upon AI-generated legal authorities without independent human verification. Consequently, maintaining professional accountability is not merely an ethical obligation but also a constitutional imperative.
Some scholars argue that increasingly autonomous AI systems may eventually require new models of legal liability because they increasingly influence legal outcomes. From this perspective, AI developers or technology providers may share responsibility where defective systems generate inaccurate legal information. While this debate is important, it does not alter the current legal position. AI lacks legal personality and cannot owe professional duties or answer disciplinary proceedings. Consequently, responsibility for legal submissions must remain with the practitioner who chooses to rely upon AI-generated outputs.
The argument that artificial intelligence should bear responsibility for its own errors is, at present, neither legally nor conceptually sustainable. Professional responsibility presupposes legal personality, the capacity to owe enforceable duties and the ability to answer for misconduct before a competent disciplinary or judicial body. Artificial intelligence satisfies none of these requirements. It cannot be admitted as a legal practitioner, owe fiduciary duties to a client, exercise independent ethical judgment or appear before the Legal Practice Council to answer allegations of professional misconduct. Responsibility for AI-generated legal work must therefore remain with the human practitioner who exercises professional judgment and decides to rely upon the technology.
This conclusion is also consistent with international developments. The Council of Europe’s European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems and their Environment identifies transparency, human oversight, accountability and respect for fundamental rights as essential principles governing the responsible use of AI within legal systems.[11] Likewise, UNESCO’s Guidelines for the Use of AI Systems in Courts and Tribunals recognise that AI should support, rather than replace, human legal decision-making.[12]These instruments reflect a growing international consensus that technological innovation should complement legal practice without weakening the ethical responsibilities that remain central to the administration of justice.
Accordingly, the question is not whether artificial intelligence should be incorporated into legal practice. The more important question is how its use should be regulated to preserve public confidence in the legal profession while allowing practitioners to benefit from technological innovation. South African law already provides a strong foundation by placing professional responsibility upon the practitioner rather than the technology. What remains is the development of more specific guidance governing the ethical use of generative AI within legal practice.
Towards a Responsible Regulatory Framework
The increasing use of AI should not discourage innovation. Properly regulated, AI can improve efficiency, reduce research costs and expand access to justice. South Africa should therefore encourage responsible AI use while ensuring that professional accountability remains with legal practitioners.[13]
However, technological innovation must operate within an ethical and regulatory framework that preserves public confidence in the legal profession. The Legal Practice Act and the existing Code of Conduct establish broad duties of honesty, competence and diligence, but they do not provide detailed guidance on the responsible use of generative artificial intelligence. As AI becomes increasingly integrated into legal practice, the Legal Practice Council should consider issuing practice directives specifically regulating its use. Such guidance could require practitioners to verify every AI-generated authority before filing court documents, prohibit the disclosure of confidential client information to unsecured AI platforms, and require continuing professional development on the ethical use of artificial intelligence.
Academic commentators have similarly argued that regulation should focus on the conduct of legal practitioners rather than the technology itself. Van Eck observes that artificial intelligence should function as an assistive tool rather than a substitute for professional judgment and warns that ignorance of AI’s limitations cannot excuse breaches of legal ethics.[14] This approach recognises that accountability is inseparable from professional status. A legal practitioner owes enforceable duties to clients, the courts and society because of admission to the profession. Artificial intelligence, by contrast, possesses neither legal personality nor moral agency and therefore cannot bear legal or disciplinary responsibility for professional misconduct.
Accordingly, South Africa should adopt a proactive regulatory approach that encourages innovation while safeguarding the integrity of legal practice. Clear ethical standards governing the use of generative AI would reduce uncertainty, promote responsible technological adoption and strengthen public confidence in the justice system. Such reforms would also ensure that the legal profession continues to fulfil its constitutional obligation to uphold the rule of law while embracing technological advancement.
Conclusion
Artificial intelligence is reshaping the legal profession by transforming the manner in which legal research, drafting and case preparation are performed. Its capacity to improve efficiency and expand access to legal services cannot be ignored. Nevertheless, recent judicial decisions demonstrate that these benefits are accompanied by significant risks where AI-generated information is accepted without independent verification. The decisions in Mata v Avianca, Inc. and Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others make it clear that legal practitioners cannot avoid responsibility by attributing errors to artificial intelligence.
This article has argued that AI cannot bear legal or professional responsibility because it lacks legal personality, enforceable duties and the capacity to exercise ethical judgment. Responsibility for AI-generated errors must therefore remain with legal practitioners, whose obligations of competence, honesty and diligence continue to apply regardless of the technological tools they employ. As artificial intelligence becomes increasingly embedded within legal practice, regulatory frameworks must evolve to promote innovation while preserving professional accountability.
The future of legal practice lies not in replacing lawyers with artificial intelligence but in ensuring that AI remains a tool subject to human oversight. While AI can improve efficiency, only legal practitioners can exercise ethical judgment and bear professional responsibility. Maintaining this distinction is essential to protecting public confidence in the administration of justice.
Bibliography
Cases
Mata v Avianca, Inc 678 F Supp 3d 443 (SDNY 2023).
Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others (7940/2024P) [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP).
Parker v Forsyth NO and Others (Regional Court, Johannesburg, Case No 1585/20, 29 June 2023).
Legislation
Constitution of the Republic of South Africa, 1996.
Legal Practice Act 28 of 2014.
Code of Conduct for all Legal Practitioners, Candidate Legal Practitioners and Juristic Entities.
Books
Susskind R, Tomorrow’s Lawyers: An Introduction to Your Future (3rd edn, Oxford University Press 2023).
Journal Articles
Dahl M and others, ‘Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models’ (2024).
Van Eck M, ‘Error 404 or an Error of Judgment? An Ethical Framework for the Use of ChatGPT in the Legal Profession’ (2024).
Reports and Guidelines
Council of Europe, European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems and their Environment (CEPEJ 2018).
UNESCO, Guidelines for the Use of AI Systems in Courts and Tribunals (2025).
[1] Mata v Avianca, Inc 678 F Supp 3d 443 (SDNY 2023); Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others (7940/2024P) [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP).
[2] Richard Susskind, Tomorrow’s Lawyers: An Introduction to Your Future (3rd edn, Oxford University Press 2023).
[3] Matthew Dahl and others, ‘Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models’ (2024); M van Eck, ‘Error 404 or an Error of Judgment? An Ethical Framework for the Use of ChatGPT in the Legal Profession’ (2024).
[4] Mata v Avianca, Inc 678 F Supp 3d 443 (SDNY 2023).
[5] Mavundla v MEC: Department of Co-operative Governance and Traditional Affairs KwaZulu-Natal and Others (7940/2024P) [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP).
[6] Parker v Forsyth NO and Others (Regional Court, Johannesburg, Case No 1585/20, 29 June 2023), discussed in Mavundla (n 5).
[7] Legal Practice Act 28 of 2014.
[8] Code of Conduct for all Legal Practitioners, Candidate Legal Practitioners and Juristic Entities.
[9] Constitution of the Republic of South Africa, 1996 s 1(c).
[10] Constitution of the Republic of South Africa, 1996 ss 34 and 165.
[11] Council of Europe, European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems and their Environment (European Commission for the Efficiency of Justice (CEPEJ), 2018).
[12] UNESCO, Guidelines for the Use of AI Systems in Courts and Tribunals (2025).
[13] Richard Susskind, Tomorrow’s Lawyers: An Introduction to Your Future (3rd edn, Oxford University Press 2023).
[14] M van Eck, ‘Error 404 or an Error of Judgment? An Ethical Framework for the Use of ChatGPT in the Legal Profession’ (2024).
