In this article, Jamie Smith KC and Isabel Barter discuss Cork and another v Mark Smith [2026] EWHC 1199 (Ch) in which ICC Judge Mullen held that the Defendant firm (the “Firm”) had misled the Court not once but twice, by (a) sending a letter to the Court which relied upon a non-existent Insolvency Rule, as a result of AI hallucination; and (b) not coming clean in a second letter to the Court which was supposed to explain the first.
The Firm had made a ‘block transfer application’ (whereby the court has the power to remove an insolvency practitioner from office) under The Insolvency (England and Wales) Rules 2016 (the “Insolvency Rules”). Specifically, the application provided for the administrator or liquidator for ‘release’ or ‘discharge’ from liability. On seeing the application, ICC Judge Mullen wrote to the Defendant firm to ask what power the Court had to order such a release in respect of liquidation.
The first letter
On 30 March 2026, the Firm wrote to the Court. It stated that the Court had an express power to release under Rule 12.37(5) of the Insolvency Rules, and after the words “which provides as follows” quoted directly from the alleged rule (the “Purported Text”).
The reference to a specific power “came as a surprise” to the Judge.[1] He checked Rule 12.37(5) of the Insolvency Rules on legislation.gov.uk website and a practitioner text, and found that the Rule 12.37(5) said nothing of the sort.[2] He also looked in vain to see whether there were any other rules with the Insolvency Rules with the Purported Text (in case the paragraph reference was wrong).[3]
The Court asked the Firm for an explanation
The Judge (via the Court office) wrote to the Firm, pointing out that Rule 12.37(5) did not say what the Firm claimed it did, and asked them to identify where in the Insolvency Rules the quoted words were to be found, and “explain why they have asserted, in terms, that a rule contains express powers that it does not?”[4]
The second letter
On 14 April 2026, the Firm wrote again to the Court (the “Second Letter”). In the Second Letter, the Firm accepted that the Purported Text did not appear in Rule 12.37(5) or elsewhere, and that the use of the phrase “which provides as follows” had “The effect of presenting that wording as the text of the rule, even though it was not intended as a direct quotation”. It suggested that the wording “was a summary conclusion, drawn from reading provisions of Rule 12.37 [going on to expand on why that was said to be the case]”.[5]
The Judge’s response
The Judge noted that the explanation was impossible to accept. The Purported Text was a quotation, not a summary. The Judge was concerned that the Firm were taking a “cavalier attitude” to material being put before the Court, and that it was likely an AI hallucination which had not been checked (and the attempt to explain it away appeared to be untruthful). He listed a hearing to determine whether any further action should be taken against the solicitors involved in the production of the First Letter and Second Letter, and (if so) what.[6]
He directed that the Firm file a witness statement addressing why the First Letter had purported to set out the text of the Insolvency Rules which did not exist and whether the Purported Text had been generated by AI, and (if the explanation differed from the second letter) why that was the case.
The evidence filed
No witness statement was provided from the unidentified junior solicitor who had had day-to-day conduct of the matter (“Lawyer A”).
The Firm provided a statement from: the senior associate supervising Lawyer A; the partner responsible for supervising the senior associate; and Deputy General Counsel (Risk and Regulatory); and the Compliance Officer who produced transcripts of the chat that Lawyer A had with AI (which ran to 59 pages).
The hearing
The Firm was represented by Paul Mitchell KC, who informed the Court that the Firm had referred itself to the SRA and would cooperate fully with any investigation that the SRA thought appropriate. He submitted that as a result, no further action need taken by the Court.
The Court’s findings
The Court set out the key parts of the Ayinde judgment, and considered the Firm’s own AI policy.
The Court concluded that the First Letter contained a misleading statement of the law, and that this had come about because:[7]
- A junior associate had used AI to research the point and the AI invented the Purported Text;[8]
- Lawyer A seemed to have almost exclusively relied upon AI to provide the answers and did not check its references, even when told to do so by the AI itself; [9]
- Lawyer A did not seem to have alerted their supervisors to this fact, or apparently tell them that AI had been used;
- Lawyer A did not seem to have alerted their supervisors to the fact that the AI had been unable to confirm the text of the rule and had told Lawyer A that, if it could not be verified, it should not be presented as a direct quote, including by removal of quotation marks. All Lawyer A had done was remove the quotation marks without altering any of the other text.
The Court noted that that the Firm had not provided the above explanation in the Second Letter. The Second Letter (which AI ultimately drafted[10]) sought to justify the first.[11]
- It was an after-the-event construction of a rationale, stating that the Purported Text was a mere summary of a number of provisions of the Insolvency Rules and should not be understood as a quotation.
- That was an unjustifiable response and an opportunity to set the record straight became a further instance of misleading information being put before the court.
The Court accepted that neither the senior associate nor the partner intended to mislead the Court.[12] Neither of them knew that AI was being used and might have expected that Lawyer A would only present text as a quote from statute if Lawyer A had identified it as an authoritative source (albeit there appeared to be a failure to supervise Lawyer A adequately).
As to the Second Letter, the senior associate should have more thoroughly investigated the reasons for an “egregious misstatement” having been placed by the Court, and should have picked up the fact that the explanation provided in the Second Letter did not hold any water.[13]
One remarkable point to note about the Second Letter was that it was the AI itself which suggested that the Firm offer an unreserved apology. Lawyer A’s initial response was that they did not think that the Firm should apologise (although an apology of sorts was eventually provided).
No further sanction
The Court was satisfied that the Firm’s own referral to the SRA was an appropriate response. In respect of the senior associate and partner, mere negligence as to the falsity of material was not sufficient to justify contempt of court proceedings.[14]
As to Lawyer A, the Court considered in detail whether contempt of court proceeding should be initiated. Despite excoriating Lawyer A’s conduct,[15] the Judge as a matter of discretion considered that that would be a disproportionate step.[16] The Court accepted Mr Mitchell KC’s submission that the SRA would be best placed to investigate what had happened.[17]
Talking points
There are two particular talking points that arise from the Cork decision.
The first is a familiar one. As the Judge emphasised in his concluding paragraph, the use of AI in no way relieves the legal professional of the ultimate responsibility of doing the requisite work with the proper care. That applies even to “a very junior legal professional”.
The second talking point concerns the need for transparency in the use of AI by one member of a legal team. There are two sides to this coin. On one side, Cork illustrates the wisdom of adopting in-house AI policies that require a team member to inform other members of when and how AI is used. The other side of the coin concerns supervision. Senior lawyers must take due steps to seek to understand what use of AI is being made by more junior members and whether that use is safe and reliable.
© Jamie Smith KC and Isabel Barter
This article is not intended as a substitute for legal advice. Advice about a given set of facts should always be taken.
[1] Para. 19
[2] Para. 20.
[3] Para. 20.
[4] Para. 21.
[5] Para. 22.
[6] Paras. 23-25 and para. 69.
[7] Para. 71.
[8] As the Judge noted, “on a large number of occasions, [the AI] is plainly wrong or, at the very least, extremely misleading”, para. 33.
[9] See e.g. paras. 40, 46, 47, 48, and the Court’s general comment at para. 49 that “Despite the repeated warnings, including the warning that the court would check the wording, and the encouraging words from the AI that checking would only take a few minutes, this does not seem to have been done. Had that check been carried out, the error would have been obvious”.
[10] See para. 74.
[11] Para. 72.
[12] See para. 73.
[13] Para. 74.
[14] Para. 82.
[15] See paras. 84-85, 87-88.
[16] Para. 89.
[17] Para. 89.
