Regulation explainerLegal Services
Generative AI and lawyers' conduct rules: what each duty requires of a firm
Generative AI does not get its own rulebook in legal practice: the existing duties of competence, confidentiality, client communication, supervision, reasonable fees and candour to the court already govern it. This page maps those duties for the US and for England and Wales, explains what the fabricated-citation cases decided, and turns each obligation into a control a firm can run.
On this page
- Why generative AI use is a conduct question, not only an IT question
- Instruments and decisions that set the conduct baseline
- Six duties side by side, with the control that answers each
- What the fabricated-citation cases actually decided
- Controls that turn the duties into daily practice
- A hypothetical mid-sized firm rolls out a drafting assistant
- Questions and answers
- Sources
Why generative AI use is a conduct question, not only an IT question
Most firms first meet generative AI as a procurement or security decision: which tool, which vendor terms, which data may go in. Regulators frame it differently. The lawyer who relies on a model's output owns that output, and the partner who supervises the lawyer owns the system of supervision. A tool approved by IT shifts neither responsibility.
In England and Wales, the Solicitors Regulation Authority has said that firms and solicitors remain responsible for AI outputs and cannot hand that responsibility to an IT team or an external provider3. In the US, the American Bar Association's ethics committee reached the same position by walking through the Model Rules one at a time1. The practical consequence is that a generative AI rollout needs a conduct owner, usually the risk or general counsel function, alongside the technology owner.
Instruments and decisions that set the conduct baseline
Model rules and regulator codes state the duties; ethics opinions, regulator reports and judgments show how they apply to generative AI.
ABA Formal Opinion 512, Generative Artificial Intelligence Tools[^1]
United States (guidance on the ABA Model Rules of Professional Conduct)Applies whenA lawyer uses a generative AI tool in client work; the opinion was issued on July 29, 2024 and interprets the Model Rules, which states adopt in their own form1.
- Competence under Model Rule 1.1 includes a reasonable understanding of the capabilities and limits of the specific tool used1.
- Informed client consent is needed before information relating to the representation goes into a self-learning tool, and boilerplate in an engagement letter is not enough1.
- Fees under Model Rule 1.5 must be reasonable; hourly billing covers time actually spent, not time the tool saved1.
SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs
England and WalesApplies whenA solicitor or other regulated individual provides legal services, whatever tools are used2.
- Do not mislead the court or others, including by allowing the acts or omissions of others (paragraph 1.4)2.
- Put forward only properly arguable submissions and draw the court's attention to relevant cases (paragraphs 2.4 and 2.7)2.
- Provide a competent service, keep skills up to date and remain accountable for work done by those you supervise (paragraphs 3.2, 3.3 and 3.5)2.
SRA Risk Outlook report: The use of artificial intelligence in the legal market
England and Wales (regulator guidance, November 2023)[^3]Applies whenA regulated firm adopts or permits AI tools in its practice3.
Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions[^4]
US District Court for the Southern District of New York (June 22, 2023)[^4]Applies whenA filing cites authorities generated by an AI tool without verification4.
R (Ayinde) v London Borough of Haringey; Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin)[^5]
England and Wales, Divisional Court of the King's Bench DivisionApplies whenFalse citations, possibly AI-generated, are placed before a court5.
- Lawyers must check AI-assisted research against authoritative sources before relying on it5.
- Those with leadership responsibilities, such as managing partners and heads of chambers, must take practical and effective measures so that everyone providing legal services understands their obligations when using AI5.
Six duties side by side, with the control that answers each
Read across a row to see the US source, the England and Wales source and the firm control that addresses both.
| Duty | US (ABA Model Rules) | England and Wales (SRA Code) | Control a firm can run |
|---|---|---|---|
| Competence | Rule 1.1: understand the tool's capabilities and limits1 | Paragraphs 3.2 and 3.3: competent service, skills kept current2 | Tool-specific training before access, refreshed when the tool changes |
| Confidentiality | Rule 1.6: informed consent before input to self-learning tools1 | Paragraph 6.3: keep client affairs confidential2 | Approved tools only, no-training vendor terms, matter-scoped access |
| Client communication | Rule 1.4: disclose AI use where it matters to the client's decisions1 | Paragraph 8.6: information clients can understand2 | A firm position on what clients are told, in engagement terms |
| Supervision | Rules 5.1 and 5.3: policies, training, vendor diligence1 | Paragraph 3.5: accountable for supervised work2 | Named supervisor per tool, usage logs reviewed, escalation route |
| Fees | Rule 1.5: reasonable fees; bill actual time1 | Paragraph 8.7: best possible information on pricing2 | Written billing policy for AI-assisted work and tool costs |
| Candour to the court | Rules 3.1 and 3.3: verify authorities before filing1 | Paragraphs 1.4, 2.4 and 2.7: do not mislead; properly arguable2 | Mandatory citation check against an authoritative source before filing |
References point to the model texts; US state rules and the SRA's other codes add to or vary these duties.
What the fabricated-citation cases actually decided
In Mata v. Avianca, the court said that using a reliable AI tool for assistance is not inherently improper; the failure was submitting non-existent opinions with fake quotations created by ChatGPT and standing by them after the court questioned them4. The court found bad faith and ordered letters to the client and to each judge falsely named as an author4. The lesson for firm design: the verification failure was compounded by a failure to escalate and correct.
The Divisional Court in Ayinde said AI tools are not capable of reliable legal research, named the authoritative sources output must be checked against, and warned of regulatory referral, wasted costs orders and, where the threshold is met, contempt5. It added that guidance alone is not enough and that it expects to ask whether leaders took effective measures5. Judicial guidance published in October 2025 likewise expands on hallucination risk6, so expect courts to ask how AI-assisted material was checked.
Controls that turn the duties into daily practice
Each item answers at least one duty in the table above. Tick them off for each tool, not once for the whole firm.
A hypothetical mid-sized firm rolls out a drafting assistant
Questions and answers
Do we have to tell clients that we use generative AI on their matters?
Not always. The ABA opinion links disclosure to the circumstances: when a client asks, when the tool's use affects decisions the client must make, or when client information goes into a self-learning tool, which needs informed consent. The SRA's AI report says firms must decide what they tell clients about AI use. Many firms settle this in engagement terms and add matter-specific consent where data handling changes.
Can a firm charge clients for the cost of an AI tool?
It depends on how the cost is treated. The ABA opinion distinguishes general overhead, which is usually absorbed, from costs tied to a specific matter, and says lawyers should not bill for time spent learning a tool they will use generally. In England and Wales, the SRA Code requires the best possible information about pricing. Write a billing policy before rollout so fee earners are not deciding case by case.
Who in a firm is responsible if an associate files a fabricated citation?
The associate who filed it, first. The Divisional Court in Ayinde also said that managing partners and others with leadership responsibility must take practical measures so everyone understands their obligations, and that it would ask whether they had. The court referred the supervising solicitor in that case to the SRA. In the US, Mata held the firm jointly responsible for the sanction. Supervision arrangements are therefore part of the exposure.
Sources
- Formal Opinion 512: Generative Artificial Intelligence Tools — American Bar Association Standing Committee on Ethics and Professional Responsibility · checked 10 October 2026
- SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs — Solicitors Regulation Authority · checked 10 October 2026
- Risk Outlook report: The use of artificial intelligence in the legal market — Solicitors Regulation Authority · checked 10 October 2026
- Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (ECF 54) — US District Court for the Southern District of New York, via CourtListener RECAP · checked 10 October 2026
- Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin) — The National Archives, Find Case Law · checked 10 October 2026
- Artificial Intelligence (AI) – Judicial Guidance (October 2025) — Courts and Tribunals Judiciary · checked 10 October 2026