AI Hallucinations in Legal Work: What the Courts and the SRA Have Said

Courts in England and Wales and the Solicitors Regulation Authority (SRA) hold the lawyer responsible for checking output from artificial intelligence (AI) tools. The Divisional Court said so in Ayinde (2025). Later decisions produced SRA referrals, a firm's self-report and a public admonishment, and the SRA's August 2026 warning notice says AI does not transfer professional responsibility. In my view the same duty covers documents conveyancers sign.

I am Chief Operating Officer of Lex Sentinel, which builds AI tools for conveyancers. This article is general information for firms regulated by the SRA in England and Wales and is not legal advice. Case details come from the published judgments and the SRA pages listed under Sources; check them before relying on or quoting them. Opinions and suggestions are marked as mine.

What is an AI hallucination?

An AI hallucination is text a language model produces that is fluent and plausible but not true. In legal work the usual forms are invented case citations, real cases cited for propositions they do not support, and quotations of statutes or rules that do not say what the output claims. The SRA's warning notice on the misuse of AI, published on 17 August 2026, names court or other documents containing false or incorrect information, including citations, as one of its two main concerns. The other is the risk to client confidentiality when information is entered into tools without safeguards.

What did the courts decide in Ayinde and Al-Haroun?

In Ayinde and Al-Haroun, the Divisional Court held that lawyers have a professional duty to check the accuracy of AI-assisted research against authoritative sources. The decision of 6 June 2025 is Ayinde v Haringey and Al-Haroun v Qatar National Bank ([2025] EWHC 1383 (Admin)), and it covered fabricated citations in two separate cases. In Al-Haroun, 18 of the 45 citations in the schedule of authorities did not exist. Many of the rest did not contain the quotations attributed to them or support the propositions cited.

The court said that generative AI tools trained on a large language model, such as ChatGPT, are not capable of conducting reliable legal research. It said that putting false material before the court can lead to a wasted costs order, referral to the regulator and, in the most serious cases, contempt proceedings. Mere negligence is not enough for contempt. In Ayinde the court said the threshold for starting contempt proceedings was met but declined to start them. In the earlier Ayinde hearing, Ritchie J had ordered the barrister and the Haringey Law Centre to pay £2,000 each in wasted costs and required referrals to the Bar Standards Board and the SRA. In Al-Haroun costs were reserved, and the court itself referred the solicitor concerned to the SRA.

Have there been later cases?

Yes: at least three decisions since Ayinde dealt with fabricated or inaccurate citations, and none concerns conveyancing. They came from the Chancery Division of the High Court (Ch), the Upper Tribunal, Immigration and Asylum Chamber (UKUT (IAC)), and the Family Court (EWFC).

In Cork v Smith ([2026] EWHC 1199 (Ch), 22 May 2026), a very junior solicitor at Pinsent Masons used the firm's AI tool, which attributed to Insolvency Rule 12.37(5) wording that does not appear in it. The wording went to the court in two letters as if it were a quotation. The firm referred itself to the SRA and agreed to pay the clients' additional costs. ICC Judge Mullen's judgment was a public admonishment of the firm and two named lawyers, and said legal professionals cannot outsource legal research or reasoning to an AI.

In UK and R (Munir) v Secretary of State for the Home Department ([2026] UKUT 81 (IAC)), promulgated on 17 November 2025 according to the National Archives page, the Upper Tribunal dealt with fabricated or inaccurate authorities in two matters. It said a supervisor who fails to ensure a junior's work is free of false cases is likely to be more culpable than a lawyer who fails to check their own, and referred one solicitor to the SRA. The amended judicial review claim form now requires confirmation that each authority exists, can be located from the citation and supports the proposition for which it is cited.

In Re A, B, C, D ([2026] EWFC 71 (B), 5 March 2026), an unregistered barrister acting in person relied on a public AI tool, and four citations or propositions were agreed to be erroneous. The judge absolved her of any intention to mislead the court but was concerned that she minimised the seriousness of doing so. The judgment quotes the Court of Appeal's remark that such tools are not an authoritative or infallible body of legal knowledge.

What does the SRA expect?

The SRA expects both the individual solicitor and the firm to remain accountable for AI-assisted work, and it is investigating reported misuse. The warning notice says the use of AI does not diminish or transfer professional responsibilities. It cites Principles 1, 2, 4, 5 and 7 of the SRA Principles and a list of paragraphs in the Codes of Conduct, and it says submissions of named case law authorities must be genuine, relevant and have a verifiable citation. In a press release the same day, the SRA said it had received 42 reports of potential AI misuse between July 2025 and July 2026, and that investigations were ongoing, covering inaccurate citations, supervision and confidentiality.

The notice does not mention conveyancing or property. The SRA's anti-money laundering sectoral risk assessment, updated in September 2026, is a separate document that treats residential conveyancing as one of the highest-risk areas and says deepfakes may increase the risk of identity fraud. That concerns AI used by fraudsters, and it is a different risk from a firm's own tool producing wrong output.

Does this matter if you don't go to court?

AI hallucinations matter to conveyancers who never go to court, in my view, because the failure is the same wherever the output ends up. A conveyancer signs reports on title, replies to enquiries and reports to lenders, and a fabricated finding in any of them is the same failure as a fabricated citation. A tool might invent a lender requirement, misdescribe a lease clause or misread a search result. The cases above arose in litigation, but in my reading the principle in them is general: the professional checks the output, and the work goes out under the professional's name, whichever tool produced it.

How can a firm reduce the risk?

A firm can reduce the risk by treating every AI output as a draft that a named person verifies against the source document. The sources I have read do not prescribe a method, so these are my own suggestions.

1. Require a reviewer to check every figure, quotation and citation against the original document, and record that they did.
2. Prefer tools that link each statement to its place in the source document.
3. Do not trust an AI tool to judge its own accuracy, which the SRA's 2023 Risk Outlook on AI advises, and I would extend that to using a second AI tool as the checker.
4. Make supervisors answerable for juniors' AI-assisted work; in Munir the Upper Tribunal said a supervisor who fails to check is likely to be more culpable.
5. Keep client confidential information out of public tools unless the safeguards described in the SRA's notice are in place.
6. Test the tool on synthetic files and keep the results.

About the author

Appan Pathmanathan is a solicitor who has worked in UK legal practice since 2001, including litigation involving money laundering and fraud, and conveyancing, and has been managing partner of a law firm. He delivers anti-money laundering and compliance training. Profile: LinkedIn.

Sources

Checked on 25 October 2026.

1. SRA, Misuse of AI warning notice (17 August 2026)
2. SRA, Responsible use of AI press release (17 August 2026)
3. SRA, Anti-money laundering sectoral risk assessment (updated September 2026)
4. SRA, Risk Outlook report: artificial intelligence in the legal market (20 November 2023)
5. [Ayinde v Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin)](https://www.judiciary.uk/judgments/ayinde-v-london-borough-of-haringey-and-al-haroun-v-qatar-national-bank/)
6. [Cork v Smith [2026] EWHC 1199 (Ch)](https://caselaw.nationalarchives.gov.uk/ewhc/ch/2026/1199)
7. [UK and R (Munir) v SSHD [2026] UKUT 81 (IAC)](https://caselaw.nationalarchives.gov.uk/ukut/iac/2026/81)
8. [Re A, B, C, D [2026] EWFC 71 (B)](https://caselaw.nationalarchives.gov.uk/ewfc/b/2026/71)

Frequently Asked Questions

What is an AI hallucination?

An AI hallucination is output from an AI model that reads as authoritative but is wrong, such as a case that does not exist or a rule quoted inaccurately.

Is a solicitor responsible for AI errors?

Yes, a solicitor remains responsible for errors in AI-assisted work. The SRA's warning notice of 17 August 2026 says the use of AI does not diminish or transfer professional responsibilities, and the Divisional Court in Ayinde held that lawyers must check AI-assisted research against authoritative sources.

What happens if fake citations reach a court?

Possible consequences include wasted costs orders, referral to the regulator and, in the most serious cases, contempt proceedings, according to the Divisional Court in Ayinde. In Cork v Smith the firm referred itself to the SRA and the judge publicly admonished it.

Does this apply to conveyancing?

Yes, in my view: the cases are litigation cases and the SRA notice does not mention conveyancing, but the principle that the professional checks the output applies to any document a conveyancer signs.

Can one AI tool check another for hallucinations?

Not reliably, in my view. The SRA's 2023 Risk Outlook advises against trusting an AI system to judge its own accuracy, and the safer course is for a person to verify the output against the source.