Key takeaways
- In Ayinde and Al-Haroun the Divisional Court dealt with fictitious authorities put before the courts, including 18 non-existent cases in a single schedule.
- The court said freely available generative AI tools are not capable of conducting reliable legal research.
- Those with leadership responsibilities, including managing partners, were told to take practical and effective measures.
- Every citation must be checked in an authoritative source, read in context and confirmed as still good law before it is used.
What happened in Ayinde and Al-Haroun
On 6 June 2025 the Divisional Court of the King's Bench Division, Dame Victoria Sharp P and Mr Justice Johnson, gave judgment in two linked referrals: R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank, reported at [2025] EWHC 1383 (Admin)[1]. Both concerned material put before the courts that cited authorities which did not exist or did not say what was claimed.
In Ayinde, five fictitious cases were cited in the grounds for judicial review[1]. In Al-Haroun, a schedule listed 45 citations, of which 18 were cases that do not exist; the court said many of the remaining cases did not contain the quoted passages or did not support the propositions for which they were cited[1].
- Cases that do not exist 18 40%
- Other citations 27 60%
Source: The National Archives (Find Case Law), 2025[1]
Show the data
| Item | Value |
|---|---|
| Cases that do not exist | 18 |
| Other citations | 27 |
The court's central message was blunt: freely available generative artificial intelligence tools are not capable of conducting reliable legal research[1]. They can produce coherent and plausible answers that are entirely wrong, cite sources that do not exist and quote passages that are not in the real source.
A message for firm leaders
The judgment did not stop at the individuals involved. It said that practical and effective measures must now be taken by those with individual leadership responsibilities, such as heads of chambers and managing partners[1]. In other words, it is not enough for a firm to rely on each lawyer knowing the risk. Leaders are expected to put guidance, training and checks in place, and to make sure they work.
That matches the regulator's position. The SRA has said firms remain responsible and accountable for outputs from AI they use[2], and the Code of Conduct requires solicitors not to mislead, or attempt to mislead, their clients, the court or others[3]. Putting a non-existent case before a court, however it got into the document, engages that duty directly.
Why hallucinations happen
A general generative AI tool is not a database. It produces text by predicting what is likely to come next, based on patterns in what it learned from. Case names, neutral citations and quotations have a recognisable shape, so the tool can produce something that looks exactly like a real authority: plausible parties, a believable year, a court abbreviation and a paragraph number. Nothing in that process checks that the case exists.
Tools connected to a curated legal database are better, because they retrieve real documents before writing. But they can still misdescribe what a case decided, attach a quotation to the wrong judgment or miss that a decision was overturned. The risk falls; it does not disappear. Clients notice the difference: 72% of legal service users in the Legal Services Board's 2026 panel research rated accuracy as the most important aspect of AI legal tools, and 87% expected tools to be kept up to date with important changes in the law[4].
Types of citation error
| Error | What it looks like | How to catch it |
|---|---|---|
| Non-existent case | Plausible name and citation that returns nothing | Search the citation and party names in an authoritative source |
| Real case, wrong proposition | The case exists but decided something else | Read the relevant part of the judgment yourself |
| Invented or altered quotation | Quoted words that are not in the judgment | Find the quoted text in the judgment, at the cited paragraph |
| Wrong citation details | Correct case, wrong year, court or paragraph | Check the full citation against the report or official transcript |
| Overturned or superseded authority | Real case that is no longer good law | Check subsequent history and later statute |
| Misattributed statute or rule | Section numbers or rule wording that do not match | Check the current version on an official legislation source |
A verification protocol
Treat AI output as a lead, not a source
Any authority suggested by a tool is a starting point for research, never a finished citation.
Find the authority yourself
Locate every case and statute in an authoritative source, such as an official judgment archive, official legislation or a recognised law report.
Read it in context
Read enough of the judgment to confirm it supports the proposition. Check that any quotation appears, word for word, at the cited paragraph.
Check it is still good law
Look at later history: appeals, overruling, amending legislation.
Record the check
Note on the file who verified each authority, where and when. For court documents, keep the list with the draft.
Second pair of eyes for court documents
Before filing, a second person spot-checks a sample of authorities, and all of them if any AI tool was used in drafting.
None of this is new. It is what careful research has always involved. The difference is that AI tools make it easy to skip the steps because the output looks finished. A protocol written down and enforced by supervisors is what turns good intentions into practice.
Measures for managing partners
- Write the protocol into the firm's AI policy and litigation procedures, and train every fee earner and trainee on it.
- Decide which tools may be used for research, if any, and block or discourage the rest on firm devices.
- Require disclosure within the team when AI has been used to draft a document going to court.
- Make verification part of supervisor sign-off for court documents, with a checklist on the file.
- Review compliance by sampling: pick filed documents each quarter and check that the record of verification exists.
- Treat a near miss as a learning point to share, not a secret.
Confidentiality matters here too. The Law Society's generative AI primer flags the risk of misuse or disclosure of confidential information[5]. A fee earner who pastes a draft skeleton argument into a free tool to tidy the citations may create a confidentiality problem as well as an accuracy one.
Documents from counsel and the other side
The same checks apply to documents your firm did not write. When you instruct counsel, the instructions can ask whether any AI tool was used to prepare the advice or skeleton, and remind counsel that every authority will be checked before filing. Most counsel will welcome the question. When a skeleton argument or letter arrives from the other side, an unfamiliar authority deserves a quick search before you spend time distinguishing it.
If you find that an opponent has cited a case that does not exist, raise it promptly and courteously with them, and with the court if the document has been filed. The court in the linked referrals made clear that putting false material before it is a serious matter, whoever drafted the document, and it is in every party's interest that problems are found early rather than at a hearing. Keep a note of what you found and when you raised it.
Example: an invented authority caught in time
Example (fictional): a trainee prepares a first draft of written submissions for an interim application and, against firm policy, asks a general AI tool for supporting authorities. Two of the five cases it suggests look entirely plausible. The supervising associate follows the protocol, searches each citation and finds that neither exists. The draft is corrected before it leaves the firm.
The firm treats the incident as a near miss. The trainee is not disciplined, but the training is repeated for the whole trainee intake with this case as the example, and the firm adds a line to its court document checklist: any authority not found by the fee earner personally in an authoritative source is removed.
If a false citation has already been filed
- Tell the supervising partner and the COLP straight away.
- Verify every other authority in the document, not just the one found.
- Correct the position with the court and the other side promptly and candidly.
- Consider your reporting obligations to the SRA and your insurer.
- Record what happened, why, and what has changed to stop it happening again.
For the wider framework, read what the SRA expects from firms using AI and writing a firm AI policy. To understand which kinds of tool carry which risks, see AI tools for legal research and drafting compared, and for a general introduction, generative AI in law firms.
Frequently asked questions
What did the court decide in Ayinde v Haringey?
In a June 2025 judgment, the Divisional Court addressed fictitious authorities put before the courts in Ayinde and Al-Haroun. It said freely available generative AI tools are not capable of reliable legal research and that those with leadership responsibilities must take practical and effective measures.
Can solicitors use AI for legal research at all?
Only as a starting point. Every authority must be found in an authoritative source, read in context and checked as still good law before it is relied on.
Are legal research tools with AI safer than general chatbots?
Usually, because they retrieve real documents from a curated source. They can still misdescribe cases or quotations, so the same verification is needed.
What should a firm do if a fake citation has been filed?
Tell the supervisor and COLP, check every other authority, correct the position with the court and other side promptly, consider reporting obligations and record what changed.
References
The National Archives (Find Case Law) (2025). R (Ayinde) v London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin) (opens in a new tab) ↑
Solicitors Regulation Authority (2023). Risk Outlook report: The use of artificial intelligence in the legal market (opens in a new tab) ↑
Solicitors Regulation Authority (2025). SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs (opens in a new tab) ↑
Legal Services Board (2026). AI in Legal Services: Research Report (LSB Public Panel) (opens in a new tab) ↑
The Law Society (2026). Generative AI: the essentials (opens in a new tab) ↑