Regulation & Ethics
Ayinde v Haringey Explained: The UK's Leading AI Hallucination Judgment
Ayinde v Haringey [2025] EWHC 1383 (Admin) explained: the fake cases, the Divisional Court's sanctions ladder, the duty on heads of chambers and how the SRA built on it.
On this page
Ayinde v Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), decided by the Divisional Court on 6 June 2025, held that freely available generative AI tools cannot conduct reliable legal research, set out a sanctions ladder from admonition to contempt and police referral, and made heads of chambers and managing partners answerable for how their lawyers use AI.
Two cases, one judgment
The Divisional Court took both matters under its Hamid jurisdiction, the High Court's inherent power to regulate the conduct of the lawyers who appear before it. Neither case was about AI on its facts. Both became about AI once somebody tried to find the cases cited.
| Ayinde v London Borough of Haringey | Al-Haroun v Qatar National Bank | |
|---|---|---|
| Underlying proceedings | A judicial review claim against the borough | A civil claim against the bank |
| Lawyers | Sarah Forey, a junior barrister; Victor Amadigwe, solicitor | Abid Hussain, solicitor |
| What was fabricated | Five authorities that do not exist | 18 of the 45 authorities cited did not exist; others did not say what was claimed |
| How it surfaced | The defendant's lawyers could not find the cases and asked for copies | The citations could not be traced |
| Explanation offered | Forey denied using generative AI; the court found her account "unpersuasive" | Hussain accepted that he had relied on research supplied by his own client, who had used AI tools |
| Outcome | Wasted costs of £2,000 plus VAT; referrals to the Bar Standards Board and the SRA | Referral to the SRA; the court held that the duty to check rests with the solicitor, whoever did the research |
Dame Victoria Sharp P and Johnson J gave a single judgment, available from the judiciary website. It runs the two stories together because the lesson is the same: a citation the lawyer has not personally verified is a citation the lawyer cannot put before a court.
What the Divisional Court held
Four holdings matter for practice, and none of them required a new rule.
Generative AI is not a research tool. The court's central finding was that "freely available generative AI tools ... are not capable of conducting reliable legal research". Such tools produce answers that read as coherent and plausible and may be entirely wrong. That is a description of the technology, not a criticism of any lawyer, and it applied in June 2025 to the consumer chatbots that most of the profession was using.
The duty to verify is personal and non-delegable. Lawyers may use AI, but anything it produces must be checked against authoritative sources before it is relied on: the law reports, BAILII, the National Archives, the databases. Hussain's reliance on his client's research did not help him. A solicitor who signs a statement citing 45 authorities is responsible for all 45.
Existing duties already cover this. The court did not ask for legislation or a new practice direction. The duty not to mislead the court, the codes of conduct of both branches and the court's own Hamid powers are enough. The same conclusion was reached in the United States, where courts apply Rule 11 and inherent powers rather than AI-specific rules, as the Mata v. Avianca line of cases shows.
Denial is worse than the error. Forey told the court she had not used generative AI and offered an account of how the five cases came to be in her grounds. The judges found it "unpersuasive". They concluded that the threshold for contempt proceedings had been met, and declined to initiate them only because she was junior and had already been reported to the Bar Standards Board, a decision the Law Society Gazette reported under the headline that the lawyers had "escaped" contempt. That framing is accurate. The court chose not to use the power, and said so.
The sanctions ladder
The most quoted passage of the judgment lists what a court can do when a lawyer places fabricated authority before it. The rungs are not alternatives; a serious case can attract several.
| Rung | What it means | Seen since Ayinde |
|---|---|---|
| Public admonition | The court names the lawyer and the failure in a published judgment | Cork v Smith [2026] EWHC 1199 (Ch): a Pinsent Masons associate's "cavalier attitude" criticised in open court (June 2026) |
| Wasted costs order | The lawyer, not the client, pays the costs the fabrication caused | £2,000 plus VAT in Ayinde; Pinsent Masons agreed to bear the extra costs in Cork v Smith |
| Striking out | The offending document, or the case, is struck out | The Austrian Supreme Court dismissed an AI-drafted appeal outright without reaching the merits (OGH 14 Os 95/25i, 7 October 2025); the Alabama Supreme Court dismissed an entire appeal (24 April 2026) |
| Referral to the regulator | BSB or SRA investigate; discipline follows the regulator's process | Ayinde and Al-Haroun referrals; Pinsent Masons referred to the SRA (June 2026); 42 AI-misuse reports to the SRA between July 2025 and July 2026 |
| Contempt proceedings | The court's own coercive power for interfering with the administration of justice | Threshold met in Ayinde; in Ontario, Ko v. Li (2025 ONSC 6785) produced a criminal-contempt referral after the lawyer confessed to lying about who drafted the factum (December 2025) |
| Referral to the police | For the case where fabricated material is placed before a court knowingly | Listed as the top rung; reserved for deliberate deception |
| Strike-off | Not a court sanction but the regulator's end point | SRA v Kumar: struck off the Register of Foreign Lawyers after using fabricated authorities in his own SDT defence, then "correcting" them with more (decided 25 August 2026) |
The ladder is worth pinning on the wall because it corrects a common misreading of the English cases. Wasted costs of £2,000 sound like a parking fine. They were the bottom of the ladder, imposed on a junior in a first-of-its-kind case. Every case since has climbed higher.
The duty that travels upward
The passage of the judgment that changed the most behaviour was addressed not to the lawyers in the room but to their principals. Heads of chambers and managing partners, the court said, must ensure that every lawyer in their organisation understands their professional and ethical obligations when using AI, and that practical and effective measures are in place to secure compliance. Burges Salmon's analysis reads this as a supervisory duty in its own right, and it is hard to disagree.
For firms with any European footprint the parallel is obvious. Article 4 of the EU AI Act has required organisations since 2 February 2025 to take measures on the AI literacy of staff who operate AI systems, with national authorities supervising it since 2 August 2026. Ayinde arrives at a similar place through the common law: the person at the top must be able to show what was done. A defensible answer looks like this.
Head of chambers and managing partner checklist after Ayinde
- A written AI policy that names the permitted tools, prohibits the use of consumer chatbots for legal research and states that every citation is verified by the responsible lawyer before filing (a policy template is on this site)
- A verification step in the filing workflow, with a record kept in the matter file, following a citation-checking protocol
- Training for every fee-earner and pupil on how generative AI produces text and why it fabricates, with attendance recorded, which also serves as the AI literacy record
- A rule that research supplied by clients, litigants in person or third parties is checked as if the firm had produced it
- A reporting route: who a junior tells, the same day, when a citation cannot be found
- A response protocol for the court and opposing party (below), agreed before it is needed
- Confidentiality rules that stop client material being pasted into public tools, since the Upper Tribunal held in UK v SSHD [2026] UKUT 81 (IAC) that doing so puts it "in the public domain"; the confidentiality checklist covers which tools train on inputs
- An annual review of the policy against the SRA Warning Notice, the Bar Council's considerations and the judiciary's guidance
From Ayinde to the SRA Warning Notice
Ayinde was not the first English case with fabricated authorities. In Bandla v SRA [2025] EWHC 1167 (Admin), decided in May 2025, fake cases were cited to the High Court against the solicitors' regulator itself (Mountford Chambers). It was, though, the first to state the duties in general terms, and everything since builds on it.
| Date | Development |
|---|---|
| May 2025 | Bandla v SRA: fictitious authorities cited against the regulator |
| 6 Jun 2025 | Ayinde and Al-Haroun: duties stated, sanctions ladder set out, principals made responsible |
| 31 Oct 2025 | Judiciary's AI guidance for judicial office holders updated: no private information in public AI tools; judges responsible for all material produced in their name (judiciary.uk) |
| 25 Nov 2025 | Bar Council's "Considerations when using ChatGPT and generative AI" reviewed |
| 2026 | BCP v A Mother [2026] EWFC 71 (B) in the Family Court; UK v SSHD [2026] UKUT 81 (IAC) in the Upper Tribunal |
| 4 Jun 2026 | Cork v Smith [2026] EWHC 1199 (Ch): an associate asked an AI tool about insolvency law and it invented "Insolvency Rule 12.37(5)"; the tool had flagged its own possible inaccuracy and the associate did not check; firm referred to the SRA (UK Human Rights Blog) |
| 17 Aug 2026 | SRA Warning Notice on the misuse of AI: 42 reports in twelve months; solicitors must understand a tool "well enough to review its output critically" |
| 25 Aug 2026 | SRA v Kumar decided by the Solicitors Disciplinary Tribunal: first strike-off for AI fabrications, reported 4 September (Legal Futures) |
| Sep 2026 | Matthew Lee's UK tracker counts 76 cases; Damien Charlotin's global database records 69 for the UK among 2,041 worldwide |
Cork v Smith deserves a second reading because it shows how the problem has moved. The associate was at a large firm, not a law centre; the tool warned that its answer might be wrong; and the fabrication was a rule number, not a case name, the kind of error a glance will not catch. The judge's phrase was a "cavalier attitude". The firm's response, immediate disclosure, bearing the costs and self-referral, is the model the SRA Warning Notice now expects.
The Warning Notice itself is the regulator's translation of Ayinde into supervisory language. It does not create a new rule any more than the Divisional Court did. It tells solicitors that competence includes understanding the tool, that confidentiality includes what happens to a prompt, and that the firm's compliance officers are expected to have thought about both. The full picture of judiciary, Bar Council and Law Society guidance sits alongside it.
When a citation is challenged: the response protocol
The sharpest lesson of Ayinde is procedural. Forey's position became untenable not when the five cases were cited but when she was asked about them. Every escalated case in the UK trackers since, up to and including the strike-off in SRA v Kumar, followed a denial, a deflection or a "corrected" filing that contained new fabrications. The protocol below is the alternative.
- Stop and pull the source within the hour. Search the neutral citation and the case name on BAILII, the National Archives caselaw service and a commercial database. Do not ask the AI tool whether the case exists; the Mata lawyers did that and were told yes.
- Treat "cannot be found" as "does not exist". There is no category of cosmetic citation error once the other side has looked. Assume the worst and work back from it.
- Tell the supervising lawyer and the head of chambers or managing partner the same day. Ayinde makes it their problem too; keeping it from them removes their chance to manage it.
- Write to the court and the opposing party. Identify every affected citation, not only the one challenged. Withdraw them. Explain, factually, how they came to be there and what has been checked since. Offer to bear the costs caused.
- Do not file a corrected document until every remaining citation has been independently verified. Kumar's "corrected" response contained further fabrications and ended his career in England.
- Record the incident. What tool, what prompt, what check was missed, what has changed. The SRA will ask; so will the insurer.
- Fix the workflow, not the person. One junior's error under deadline is a system that let unread citations reach a filing. Litigators who use AI for drafting should treat the verification step as part of the drafting, not an afterthought.
Where this leaves you
Ayinde is short, readable and free, and every barrister, solicitor and trainee who touches a court document should read it once in full. Three things follow from it as of September 2026. First, generative AI can be used for drafting, summarising and structuring, and cannot be used as a source of authority; the tool is not the problem, unverified reliance is. Second, the principal is on the hook: a head of chambers or managing partner who cannot show a policy, training and a verification step has no answer if the Hamid jurisdiction is invoked. Third, the response decides the outcome, as the full sanctions casebook from both sides of the Atlantic confirms. The AI-generated citation cost £2,000; the denial nearly cost a career.
Module 6 of the AI Lab for Lawyers course works through Ayinde, Cork v Smith and the SRA notice alongside the verification routine that stops a chambers or firm becoming the next entry in the tracker.
Frequently asked questions
What is Ayinde v Haringey about?
Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin) were heard together by the Divisional Court on 6 June 2025 because lawyers in both cases had cited authorities that did not exist: five in Ayinde and 18 of 45 in Al-Haroun. The court used the case to state the profession's duties when using generative AI and the sanctions available for breaching them.
What sanctions did the lawyers in Ayinde and Al-Haroun receive?
In Ayinde, wasted costs of £2,000 plus VAT were ordered against the barrister and the solicitors, and both were referred to their regulators, the Bar Standards Board and the Solicitors Regulation Authority. In Al-Haroun the solicitor was referred to the SRA. The court found the contempt threshold met for the barrister but did not initiate proceedings given her juniority and the existing referral.
Did the court ban barristers and solicitors from using AI?
No. The judgment accepts that AI is a powerful tool with a legitimate place in legal practice. What it prohibits is relying on generative AI output for legal research without checking it against authoritative sources such as the law reports, BAILII or the National Archives. The duty to verify sits with the individual lawyer and cannot be delegated to a client, a junior or the tool itself.
What does Ayinde mean for heads of chambers and managing partners?
The court addressed them directly: those with leadership responsibilities must ensure that every lawyer in the organisation understands their professional and ethical duties when using AI, and that practical and effective measures are in place to secure compliance. In practice that means a written policy, training with a record, a verification step before filing and a reporting route when something goes wrong.
How does the SRA Warning Notice of 2026 relate to Ayinde?
The Warning Notice on the misuse of AI, published on 17 August 2026, follows Ayinde's logic. The SRA reported 42 AI-misuse reports between July 2025 and July 2026, cited the later cases BCP v A Mother, Cork v Smith and UK v SSHD, and stated that solicitors must understand any tool well enough to review its output critically. Breaches are treated under existing conduct rules, not new ones.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Regulation & Ethics
The SRA's 2026 Warning Notice on AI Misuse: What UK Solicitors Must Know
The SRA AI warning notice of 17 August 2026 followed 42 misuse reports in a year. Solicitors must understand a tool well enough to review its output critically.
8 min read
Regulation & Ethics
UK Judiciary, Bar Council and Law Society AI Guidance: The 2026 Picture
UK judiciary AI guidance (Oct 2025), Bar Council considerations (Nov 2025) and Law Society material share one rule: whoever signs is responsible for every word.
8 min read
Regulation & Ethics
Mata v. Avianca, Three Years On: What the First AI Sanction Still Teaches
Mata v. Avianca (S.D.N.Y. 2023) cost two lawyers $5,000 for six fake ChatGPT cases. What went wrong, why asking AI to verify AI fails, and what changed since.
9 min read
Regulation & Ethics
AI Hallucination Sanctions: The Cases Every Lawyer Should Know (2023–2026)
AI hallucination cases reached 2,041 in Charlotin's database by September 2026, 811 involving lawyers. Every sanction from Mata to Couvrette's $110,204, with lessons.
12 min read
Playbooks
How to Verify AI Legal Research: A Citation-Checking Protocol
Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.
11 min read
Guides
AI Literacy for Law Firms: What to Teach and What Article 4 Expects
AI literacy for law firms under Article 4 of the EU AI Act: an obligation of effort, no certificates, keep a record. A curriculum by role and a training record.
11 min read