Can UK solicitors use AI?

Yes, solicitors can use AI safely for the administrative work around their matters, but not for legal research or anything that states what the law is, because the courts in England and Wales have now dealt with several cases of invented citations and the SRA published a warning notice about it on 17 August 2026.

The case to know is R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin), handed down on 6 June 2025 by Dame Victoria Sharp P and Johnson J. In the first case, counsel put five non-existent authorities before the High Court. In the second, heard alongside it, forty-five citations were relied on and eighteen of them didn't exist - and one of the invented cases was attributed to the very judge hearing the application.

Both were referred to a Divisional Court under its Hamid jurisdiction, the court's inherent power to enforce the duties lawyers owe it. If you run a small practice and read one thing about AI this year, it's worth making it that judgment.

Why AI invents citations in the first place

It helps to know what these tools actually do. A chatbot like ChatGPT runs on a large language model, which works a bit like a very well-read autocomplete: it predicts what words are likely to come next, based on an enormous amount of text it has read. That makes it good at writing something that sounds like a case citation. It doesn't mean the case exists. It isn't looking anything up unless it's been specifically connected to a reliable source, and even then it can get things wrong.

What the court actually did

The court listed the powers it has when lawyers put false material before it: public admonition, a costs order, a wasted costs order, striking out a case, referral to a regulator, contempt proceedings, and referral to the police. A judge had already ordered the barrister and the law centre to pay £2,000 each in wasted costs - costs a lawyer is made to pay personally because of how they handled a case - and referred both to the Bar Standards Board and the SRA. The Divisional Court found the threshold for contempt proceedings was met, decided not to begin them for reasons specific to that barrister's junior status and supervision, and then said the decision "is not a precedent" and that "lawyers who do not comply with their professional obligations in this respect risk severe sanction."

Two lines matter even more than the sanctions. The first: "Freely available generative artificial intelligence tools, trained on a large language model such as ChatGPT are not capable of conducting reliable legal research." The second, about the solicitor in the second case who had relied on his client's own research: "A lawyer is not entitled to rely on their lay client for the accuracy of citations of authority or quotations that are contained in documents put before the court by the lawyer."

The court also put responsibility on the people who run firms: "practical and effective measures must now be taken by those within the legal profession with individual leadership responsibilities (such as heads of chambers and managing partners)". In future Hamid hearings, the profession can expect the court to ask whether those responsibilities have been met. In a firm of six, that's likely to be you.

It didn't stop there

The Upper Tribunal dealt with the same problem in UK and R (Munir) v Secretary of State for the Home Department [2026] UKUT 00081 (IAC), decided in November 2025 and now a reported decision, whose headnote states that a legal professional who delegates work to another fee earner "remains responsible for the supervision of their work and for ensuring its accuracy."

Then, in Cork v Smith [2026] EWHC 1199 (Ch), ICC Judge Mullen dealt with letters sent to the court by a large firm containing the supposed text of an insolvency rule that the AI had simply made up. A recently admitted associate had used a tool the firm was piloting, and the supervising solicitor accepted she had failed to supervise and check the drafts properly. The judge's conclusion is worth keeping in mind: legal professionals "cannot outsource the process of legal research or of legal reasoning to an AI."

That case adds something important. It isn't only case names that get invented. The wording of a rule, a section or a limitation period - anything that looks like a quotation - can be made up entirely.

What the SRA's warning notice says

On 17 August 2026 the SRA published a warning notice on the misuse of AI. It says the regulator has received reports of potential Code of Conduct breaches from senior members of the judiciary, and that several solicitors have self-reported after relying on AI tools that produced inaccurate content.

Its central sentence answers the obvious question before anyone asks it: "AI has no separate legal personality; solicitors and regulated individuals who use AI in the course of delivering legal services remain accountable for their work and outputs, regardless of how that work has been prepared."

It links that to provisions you'll already know: Principles 1, 2, 4, 5 and 7, and Code paragraphs including 1.4 on misleading the court, 2.4 on properly arguable assertions, 3.2, 3.5 and 3.6 on competence and supervision, and 6.3 on confidentiality. On citations it's specific: you should be assured that named authorities "are genuine, relevant, have a verifiable citation, and advance the arguments that are being put forward in your documents."

Where AI genuinely helps a small firm

The judiciary's own AI guidance for judicial office holders, refreshed on 31 October 2025, is the most useful short document on this, because it separates the two lists clearly.

Potentially useful: summarising large bodies of text, taking care that the summary is accurate; help with presentations; and administrative work, including composing, summarising and prioritising emails, transcribing and summarising meetings, and composing memoranda. Not recommended: legal research, because these tools "are a poor way of conducting research to find new information you cannot verify independently", and legal analysis, because current public chatbots "do not produce convincing analysis or reasoning."

That fits a high street practice closely. The value is in the administrative load around matters, not in the law itself: onboarding chasers nobody has time to send, appointment arrangements, a first draft of routine correspondence that a fee earner then edits, a summary of a long bundle that a fee earner then checks against the bundle, and a file note from a meeting recording. It's unglamorous work, and it's where a lot of unbilled hours go in a firm of two to ten.

If onboarding chasers and appointment arrangements are taking up fee earners' time, we're happy to set up an automation that sends the reminders and books the slots, with nothing touching the legal work.

The judicial guidance also states the principle that decides everything else: judicial office holders are "personally responsible for material which is produced in their name." The same principle applies to solicitors.

Client data is the other half of the problem

In the Upper Tribunal case, one of the lawyers involved had put client emails and Home Office decision letters into ChatGPT. The tribunal said that uploading confidential documents into an open-source AI tool such as ChatGPT "is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege", and that a firm doing so should tell its regulator and consult the ICO. It also noted that closed tools, which don't place information in the public domain, are available for tasks such as summarising.

One correction to the tribunal's wording: ChatGPT isn't open-source, meaning its code isn't published for anyone to use and change. The tribunal's point is about putting client documents into public AI tools, and it didn't look at business plans that come with a data processing agreement, the contract setting out what the provider may and may not do with your data.

The SRA puts it more carefully and, for a small firm, more usefully. It says both free and paid AI systems may carry risks, because whether information is stored, kept or used to improve the tool depends on the provider's terms, settings and technical set-up. Client information should only go into a system where the contractual, technical and organisational safeguards are actually in place.

Working out whether they are, for the specific plan you're on, is fiddlier than most people expect, and not something to assume from a marketing page. We've written separately about whether AI tools train on your business data, which covers how that differs between consumer and business plans. Once you've chosen a plan whose data terms you're comfortable with, we can help set up the admin automations on it.

Insurance, and the rules still being written

The Law Society's guide to generative AI, last updated in June 2026 and aimed particularly at small and medium-sized firms, tells you to assess liability and insurance cover for any tool, and to "speak to your insurer to determine whether the usage of the tool will be covered under your existing policy." That's a phone call rather than a project.

Our sources contradict each other on one point. The same Law Society guide says there's currently no SRA guidance on the use of AI in court documents. That was true when it was last updated, but it isn't any more, because the warning notice came two months later. Where guidance moves this quickly, it's best to check the regulator's own page rather than a summary of it.

On procedure, the Civil Justice Council working group chaired by Birss LJ is considering whether rules are needed for AI use in preparing court documents. Its interim position records a high degree of agreement among respondents that no extra AI-specific requirements are currently needed for pleadings and advocacy documents, because professional responsibility frameworks are considered sufficient. Witness statements are still under consideration. For now, the existing rules are the rules.

What we'd suggest

Decide which AI tool the firm uses and on which plan, and check what that plan does with what you type into it.

Put it in writing, on one page, that nothing stating what the law is - no citation, no quotation from a rule or statute, no legal proposition - leaves the firm unless a person has opened the authority itself.

Extend your existing supervision to AI-assisted work explicitly, because both the Upper Tribunal case and Cork v Smith turned on supervision rather than on the technology. And ring your insurer.

Then use AI for the administrative load, where the hours actually go and where a mistake is far easier to put right. If you'd rather not set that up yourself, there's more on what we set up for small businesses, and we're glad to automate the routine admin for you.

Sources

• R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin) - The National Archives, Find Case Law
• UK and R (Munir) v Secretary of State for the Home Department [2026] UKUT 00081 (IAC) - The National Archives, Find Case Law
• Cork v Smith [2026] EWHC 1199 (Ch) - The National Archives, Find Case Law
• Misuse of AI: warning notice, 17 August 2026 - Solicitors Regulation Authority
• Artificial Intelligence (AI) Guidance for Judicial Office Holders, 31 October 2025 - Courts and Tribunals Judiciary
• Generative AI - the essentials - The Law Society
• Use of AI in preparing court documents - Civil Justice Council

Checked on 29 September 2026. Regulatory guidance and the case law on this are moving quickly, and the Civil Justice Council's work is not finished, so check the links rather than trusting the date on this page.