← All guidelines·Guideline 2 of 8
The brief belongs to the counsel who accepted it. AI may assist with the work; it may not do the thinking. No argument, summary or answer is adopted that counsel cannot explain and defend as their own.
Red · Must An existing duty. Breach may be misconduct.
The rules
“Counsel shall personally attend to all of the work involved in the briefs accepted by them”
LPC Code of Conduct, para 28.4
Paragraph 28.4 goes on to preserve the long-standing practice of the devil: by agreement, another counsel may undertake research work in a particular brief for a fee paid by the counsel on brief. Assistance has always been permitted. The brief, and the responsibility for it, stay with the counsel who accepted it.
“retain the independence necessary to enable them to give their clients or employers unbiased advice;”
LPC Code of Conduct, para 3.9
“use their best efforts to carry out work in a competent and timely manner and not take on work which they do not reasonably believe they will be able to carry out in that manner;”
LPC Code of Conduct, para 3.11
The GCB Uniform Rules treat the devil in the same way. Practising advocates “should retain their professional independence” (rule 4.26.1), and the devil may not take an active part in the conduct of the case in court without the attorney’s prior consent (rule 4.26.2).
What the guideline requires
AI can find material, summarise a record, propose a structure and produce a draft. It is treated as a devil whose work is checked before it is used. The judgment calls remain counsel’s own: which points to take and which to leave, what the authorities hold, and what the evidence proves.
The test is a practical one. If the court asks where a proposition comes from, or why a point was taken, counsel must be able to answer from their own knowledge of the papers and the law. If the only answer is that the tool produced it, the work is not yet counsel’s own.
What the courts have said
The Divisional Court in England held that lawyers who use AI for legal research have a professional duty to check its accuracy against authoritative sources before using it to advise clients or before a court (Ayinde at [7]).
In FJL v TGO, Opperman J, reviewing how AI is treated across four jurisdictions, drew the distinction between AI as an assistant and AI as a decision-maker, and said of the first:
“Responsibility for accuracy in all respects rests with the author.”
FJL v TGO, Opperman J at [48]
The earliest South African case, Parker v Forsyth, put the point in a passage later quoted in Mavundla:
“Courts expect lawyers to bring a legally-independent and questioning mind to bear on, especially, novel legal matters, and certainly not to merely repeat in parrot-fashion, the unverified research of a chatbot.”
Parker v Forsyth NO at [90], quoted in Mavundla at [42]
A specialist legal tool does not alter the position. In Northbound, the incorrect citations came from a subscription tool that claimed to be trained exclusively on South African judgments and legislation (at [89]).
Green · Method The checks
- Decide the argument first. The points, and their order, are chosen before AI is asked to draft.
- Read the key sources. The principal authorities and the critical parts of the record are read by counsel, not summarised for counsel.
- Test each step. Counsel can explain why each proposition follows from the last.
- Know the other side’s best answer. AI may be asked to state it; counsel decides how it is met.
- Cut what cannot be defended. A point that could not be argued on one’s feet does not go in.
Guide 7 shows what checking an AI-produced index and chronology involves in practice; Guide 9 covers the draft itself.
Other jurisdictions
In England and Wales, a barrister is personally responsible for their own professional work, including when using AI tools; the use of AI amounts to outsourcing under rC86, and inaccuracies introduced through it remain the barrister’s responsibility (Bar Standards Board, Guidance on the use of AI and other technologies, valid from 18 May 2026). In Victoria, irrespective of the AI tools used, lawyers must exercise oversight and verify the accuracy and suitability of the information an AI system provides (Practice Note SC Gen 25, para 6.5). In New Zealand, a lawyer is not absolved from responsibility for legal advice or defects in an end product because it is derived from generative AI (New Zealand Law Society, Lawyers and generative AI, March 2024). In Ontario, generative AI is described as a valuable tool but not a substitute for the licensee’s own professional judgement (Law Society of Ontario, Generative AI: your professional obligations, practice tip 1). The international comparison has the detail.
Sources on file: LPC Code of Conduct (GG 42337, 29 March 2019), paras 3.9, 3.11 and 28.4. GCB Uniform Rules of Professional Ethics (as updated July 2024), rules 4.26.1 and 4.26.2. R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin), para [7]. FJL v TGO (2025/220239) (GJ, 31 July 2026), Opperman J concurring, para [48]. Parker v Forsyth NNO [2023] ZAGPRD 1 (29 June 2023), para [90]. Mavundla v MEC: Department of Co-operative Government and Traditional Affairs, KwaZulu-Natal [2025] ZAKZPHC 2; 2025 (3) SA 534 (KZP), para [42]. Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator [2025] ZAGPJHC 661, para [89]. Bar Standards Board, Guidance on the use of AI and other technologies (May 2026). Supreme Court of Victoria, Practice Note SC Gen 25, para 6.5. New Zealand Law Society, Lawyers and generative AI (March 2024). Law Society of Ontario, Generative AI: your professional obligations (2024), practice tip 1.
Status. A proposed guideline, put forward by Greg Fourie SC for discussion. It has not been adopted by the Legal Practice Council, the General Council of the Bar or any society of advocates. Read the Code and the judgments themselves before relying on them.