Part of the AI Regulation News hub.
An insolvency judge found that a letter from Pinsent Masons quoted a provision of the Insolvency Rules that does not exist, and that a junior solicitor had produced it using AI without her supervisor's knowledge
The judge went looking for the rule the firm had quoted. It was not in rule 12.37, not elsewhere in that chapter, and not anywhere in the Insolvency Rules 2016. What followed produced 59 pages of chat transcripts, the firm's AI policy in evidence, and a judgment the firm had to send to its regulator within two working days.
Bottom line: The court required Pinsent Masons to provide a copy of the judgment to the Solicitors Regulation Authority within two working days, so the SRA could consider it as part of an investigation it was already conducting. The judge did not need to make a referral himself because the SRA already had the details of each lawyer involved.
Who this affects: Law firm supervisors and heads of department, anyone signing off correspondence to a court that summarises statutory provisions, firms running AI pilots, and junior lawyers using AI for research.
Issue date: Hearing 18 May 2026. Judgment handed down remotely at 10.30am on 22 May 2026.
What changed: No new rule. The judgment applies the guidance of the President of the King's Bench Division in Ayinde to a firm setting, and shows what the court does when the first explanation offered does not disclose that AI was involved.
Analysis: The costliest step was not the hallucination. The judge says expressly that had the true reason for the wording been squarely set out and an appropriate apology offered, he might have concluded there was no harm done and required nothing beyond the firm meeting its clients' costs for that letter. What escalated it was an initial explanation that presented the text as a summary conclusion drawn from the rules.
Primary sources: Judgment, [2026] EWHC 1199 (Ch), Find Case Law
- Case
- (1) Anthony Malcolm Cork (2) Stephen Robert Leslie Cork v Mark Smith, in the matter of an Office-Holder and in the matter of the Insolvency Act 1986
- Neutral citation
- [2026] EWHC 1199 (Ch)
- Court
- High Court of Justice, Business and Property Courts of England and Wales, Insolvency and Companies List (ChD), Rolls Building
- Judge
- ICC Judge Mullen
- Representation
- Mr Andrew Brown, instructed by Irwin Mitchell LLP, for the Applicants. The Respondent did not appear and was not represented. Mr Paul Mitchell KC, instructed by Clyde & Co LLP, for Pinsent Masons LLP.
- Jurisdiction
- England and Wales
- Provision at issue
- Insolvency (England and Wales) Rules 2016, rule 12.37, on block transfer orders
- Status
- Judgment handed down 22 May 2026. Pinsent Masons required to provide the judgment to the SRA within two working days; counsel asked to lodge a minute of order.
- Bindingness
- A first-instance judgment. Binding between the parties; of persuasive weight, and notable for applying the Ayinde guidance in a firm-supervision setting.
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://caselaw.nationalarchives.gov.uk/ewhc/ch/2026/1199
How the court found it
The application itself was routine. A block transfer application under rule 12.37 of the Insolvency (England and Wales) Rules 2016, to move a small number of administrations and voluntary liquidations because the office-holder was changing firms. Such applications, the judgment notes, are almost invariably determined without a hearing.
The draft order also sought the release of the outgoing office-holder. Because there is no obvious express power to grant release in these circumstances, the court asked for an explanation, and the firm replied by letter dated 30 March 2026.
That letter set out wording in italics, which the judgment calls the Purported Text, introduced by the phrase which provides as follows, presenting it as the text of rule 12.37(5).
The judge checked. The words are not in rule 12.37(5). He checked the rest of that chapter of the rules; they were not there either. He then ran a word search for phrases within the Purported Text across the Insolvency Rules 2016 and could not find them at all. His stated first thought was that this was an artificial intelligence hallucination arising during the firm's research, a term the judgment pauses to define as where AI generates false or misleading information which it presents as fact.
The first explanation, and why it mattered
The court office put the point to the firm by email on 14 April 2026, quoting what rule 12.37(5) actually says.
The firm's reply accepted that the words do not appear in rule 12.37(5), elsewhere in rule 12.37, or in the Insolvency Rules 2016, and accepted that the phrase which provides as follows had the effect of presenting the wording as the text of the rule even though it was not intended as a direct quotation. It apologised for the confusion and confirmed there had been no intention to mislead.
The explanation offered for the wording, though, was that it was a summary conclusion drawn from reading provisions of rule 12.37, in particular rule 12.37(1)(e) as a catch-all discretionary power. The judgment records the judge's own view that rule 12.37(1)(e) does not provide an obvious answer, since rule 12.37(1) merely sets out what a block transfer application may ask for.
That framing is what the judge later identifies as the aggravating step, because it did not disclose that AI had been used at all.
What the evidence eventually showed
The picture that emerged from the witness statements is a supervision failure rather than a rogue actor.
Ms Samantha Poulton, a senior associate in the restructuring department admitted in 2018, had day-to-day conduct of the matter with a more junior associate identified in the judgment only as Lawyer A, recently admitted to the Roll of Solicitors. Ms Poulton's evidence was that she was aware Lawyer A used AI from time to time, but was unaware that AI was being used on this application, that it did not occur to her AI had been used here, and that she was at the time unaware of AI hallucinations. She is now undertaking a course to improve her understanding of AI, and accepts that she failed to supervise Lawyer A and to check the drafts properly.
The disclosure was substantial. The firm produced transcripts of Lawyer A's chats with the AI running to some 59 pages, and exhibited the firm's AI use policy as it stood at the time. A partner who chairs the firm's Innovation and Change Board provided a letter explaining the firm's AI pilot and the further safeguards since put in place.
Lawyer A did not provide a witness statement. The firm explained that, given their junior status and the firm's duties to its employee, and in view of the involvement of the two more senior solicitors in preparing the letters, Lawyer A had not been named, offering to arrange evidence if the court wished. The judgment refers to them throughout as Lawyer A.
The judge's reasoning on responsibility
The judgment sets out the guidance of the President of the King's Bench Division in Ayinde, including the range of powers available where false material is placed before the court: public admonition, a costs order, a wasted costs order, striking out a case, referral to a regulator, contempt proceedings and referral to the police, with the response depending on the circumstances, on whether an immediate, full and truthful explanation is given, and on steps taken to mitigate.
It also quotes the President's direction to those with leadership responsibility, that practical and effective measures must be taken so that every individual providing legal services understands and complies with their duties to the court when using AI, and that in future Hamid hearings the profession can expect the court to inquire whether those leadership responsibilities have been fulfilled.
Applying that, ICC Judge Mullen is careful about mitigation and firm about the principle. He accepts the solicitors are experienced practitioners doing technical work under pressure, and that Lawyer A was under pressure too, but holds that none of that excuses a failure to check the accuracy of material placed before the court.
His central statement is that legal professionals bear ultimate responsibility for their work and cannot outsource the process of legal research or of legal reasoning to an AI. It is a tool to be used with caution. AI has the potential to be wholly unreliable, and while it may provide a jumping off point, it does not at present do away with the need for proper research and thought on the part of a legal professional, even a very junior one.
The supervision lesson, stated by the court
The judgment contains an unusually specific finding about what should have happened after the error was flagged. The judge says that the dangers of AI should be well known by now and that, at the very least, receipt of his response to the 30 March letter should have alerted Ms Poulton to the risk that Lawyer A had used AI, which had hallucinated the text of rule 12.37(5), and should have prompted her to investigate the matter fully.
That is a duty attaching to the moment a court queries a citation, not merely to the moment of drafting. It suggests a firm's AI response protocol needs a step that most do not have: when a court or an opponent challenges a quoted provision, treat AI use as a live hypothesis and check the drafting history before answering.
On regulatory consequences, the judge was satisfied that the SRA already had the details of each of the lawyers involved and that he did not need to make a referral himself. He required Pinsent Masons to provide a copy of the judgment to the SRA within two working days of hand-down so that it could be considered as part of the investigation, and asked counsel for the firm to lodge a minute of order.
The transferable lesson is about the second step rather than the first. The court says in terms that had the true reason been squarely set out with an appropriate apology, the likely outcome was no further action beyond the firm bearing its clients' costs for the offending letter. What turned an embarrassing error into a reported judgment and a regulatory submission was an explanation that recharacterised hallucinated text as a summary conclusion. Two practical items follow: a supervisor who knows a junior sometimes uses AI cannot treat that as background knowledge when a court queries a quoted provision, and any firm AI policy should specify who investigates the drafting history, and how fast, once a citation is challenged. Note also the disclosure burden this case sets: 59 pages of chat transcripts and the firm's contemporaneous AI policy were put before the court.
Source File
https://caselaw.nationalarchives.gov.uk/ewhc/ch/2026/1199
Open the judgment and confirm four things: the neutral citation [2026] EWHC 1199 (Ch) and the date 22/05/2026 in the header, with ICC Judge Mullen named; the appearance line showing Mr Paul Mitchell KC instructed by Clyde & Co LLP for Pinsent Masons LLP; the passage describing the Purported Text and the judge's word search of the Insolvency Rules 2016; and the Conclusion at paragraphs 94 to 96, including the statement that legal professionals cannot outsource legal research or legal reasoning to an AI.
legal professionals bear ultimate responsibility for their work and cannot outsource the process of legal research or of legal reasoning to an AI. It is a tool to be used with caution. AI has the potential to be wholly unreliable. Cork v Smith, [2026] EWHC 1199 (Ch) at [95]
FAQ
What exactly was wrong with the letter?
It set out wording, introduced by the phrase 'which provides as follows', presented as the text of rule 12.37(5) of the Insolvency (England and Wales) Rules 2016. The judge checked and found those words are not in rule 12.37(5), not elsewhere in rule 12.37, and, on a word search, not anywhere in the Rules.
Was AI actually used, or was that only the judge's suspicion?
It was used. The judge's first thought was an AI hallucination, and the evidence bore that out: a junior associate, identified only as Lawyer A, had used AI, and the firm produced transcripts of those chats running to some 59 pages together with its AI use policy at the time.
Was anyone referred to the regulator?
The judge did not need to make a referral himself, being satisfied that the SRA already had the details of each lawyer involved. He required Pinsent Masons to provide a copy of the judgment to the SRA within two working days so it could be considered as part of that investigation.
Why did the court treat this as serious if the firm apologised?
Because the first explanation described the wording as a summary conclusion drawn from the rules rather than disclosing AI use. The judge said that had the true reason been squarely set out with an appropriate apology, he might have required nothing beyond the firm meeting its clients' costs for that letter.
What did the court say about supervision?
That receipt of the court's response to the letter should at the very least have alerted the supervising senior associate to the risk that the junior had used AI which hallucinated the rule, and should have prompted a full investigation. The supervisor accepted she had failed to supervise and to check the drafts properly.
Was the junior lawyer named?
No. The judgment refers to them only as Lawyer A. The firm explained that given their junior status, its duties to its employee, and the involvement of two more senior solicitors in preparing the letters, Lawyer A had not been named, and offered to arrange evidence from them if the court wished.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.