Cork v Smith: AI Hallucinations, Supervision Failures and the Court’s Firm Reminder of Professional Duties

The Chancery Division’s judgment in Cork v Smith is already being recognised as a defining early case on the use of artificial intelligence in legal practice. But the decision is not simply a warning about technology. It is a broader and more pointed reminder of the legal profession’s core obligations: to act with honesty, to verify the accuracy of material placed before the court and to supervise junior colleagues with diligence. Pinsent Masons LLP, along with two of its solicitors, received a public admonishment after submitting two misleading letters to the court: one containing fabricated statutory text generated by AI; and the second offering an explanation the court found to be unjustifiable. The matter has now been referred to the Solicitors Regulation Authority, underscoring the seriousness of the breaches.

A routine application that went badly wrong

The events unfolded during what should have been a straightforward block transfer application under Insolvency Rule 12.37. The application sought the removal of an insolvency practitioner and the appointment of a replacement. The draft order included a provision granting the outgoing practitioner his “release” without the need to apply to the Secretary of State. This immediately caught the court’s attention. While administrators have an express statutory route to discharge, liquidators do not. The court therefore asked Pinsent Masons to identify the legal basis for including such a release.

In response, the firm submitted a letter dated 30 March 2026 asserting that Insolvency Rule 12.37(5) provided the necessary authority. The letter quoted what purported to be the text of the rule. The difficulty was that the quoted text did not exist. Rule 12.37(5) deals with entirely different subject matter. The “statutory wording” had been generated by a freely available AI tool used by a junior associate, who failed to verify its accuracy despite the tool repeatedly warning that its output must be checked.

The second letter: a missed opportunity to correct the record

When the court pointed out the discrepancy, Pinsent Masons sent a second letter on 14 April 2026. Instead of candidly acknowledging the mistake, the firm claimed that the earlier wording was not intended as a quotation but as a “summary conclusion” drawn from various provisions. The court rejected this explanation outright. The first letter had clearly presented the text as a direct statutory extract, and the second letter was, in the court’s view, an unjustifiable attempt to rationalise the earlier mistake rather than a straightforward correction.

This second misstep was particularly troubling to the court. It demonstrated not only a failure to verify the original material but also a failure to take responsibility once the error was identified. The court emphasised that honesty and transparency are essential, especially when correcting mistakes that have already misled the court.

Judicial guidance on AI: the Ayinde principles

In analysing the issue, the court relied heavily on the principles set out in R (Ayinde) v London Borough of Haringey, which provided early judicial guidance on the professional use of artificial intelligence. The court reiterated that generative AI tools cannot be relied upon to conduct legal research and are prone to producing confident but entirely false assertions. Lawyers who use such tools must verify their output against authoritative sources before deploying it professionally. That duty extends beyond the individual who uses the tool; supervisors and senior lawyers who rely on work prepared by others must also ensure that the material is accurate.

The court stressed that these duties apply regardless of the nature of the application. Block transfer applications are typically determined on paper without hearings or submissions, which makes the court particularly vulnerable to being misled if solicitors do not exercise proper care. The administrative nature of the application did not diminish the seriousness of the errors.

Supervision failures and professional duties

The court accepted that neither senior associate, Ms Samantha Poulton, nor partner, Mr Steven Cottee, intended to mislead the court. However, it found that both had failed to supervise the junior associate adequately and had not checked the accuracy of the material submitted. These failures amounted to breaches of their professional duties under the SRA Code of Conduct, including the duties not to mislead the court, to act competently and to take responsibility for supervised work.

The judgment makes clear that supervision is not a box‑ticking exercise. Senior lawyers cannot assume that junior work is accurate simply because it appears polished or confident, particularly in an era where AI tools can generate plausible but entirely fictitious text. The court was especially troubled by the fact that it had been misled twice: first by the fabricated statutory text and then by the attempt to explain it away.

Why the court referred the matter to the SRA

Although the court considered whether contempt proceedings should be brought against the junior associate, it ultimately concluded that such a step would be disproportionate. The associate was inexperienced, the harm was limited and the regulatory framework provided a more appropriate route for addressing the misconduct. However, the court was clear that public admonishment alone was insufficient. The matter was therefore referred to the SRA to investigate potential breaches of professional duties, including the duty not to mislead and the duty not to waste court time.

Pinsent Masons had already self‑reported and agreed to meet the costs incurred by their former clients, including the costs of instructing alternative solicitors. While this was appropriate, it did not displace the need for regulatory scrutiny.

Lessons for the profession

The judgment carries important lessons for legal practitioners. First, AI is not a substitute for legal research. It can assist with drafting, summarising and ideation, but it cannot be trusted to produce authoritative legal analysis. Any lawyer who relies on AI without verification risks breaching their duties. Second, supervision must be active and robust. Senior lawyers must ensure that junior colleagues understand the limits of AI tools and the importance of verifying their output. Third, transparency with the court is paramount. When an error is discovered, the only acceptable response is to correct it promptly and fully. Attempts to rationalise or minimise mistakes will only compound the problem.

Finally, the case is a reminder that even administrative applications require the same level of rigour and honesty as contentious litigation. The court’s reliance on written submissions makes accuracy all the more essential.

Litigants in person

For those thinking of acting as a litigant in person, the case highlights the risks of relying on generative AI tools for legal research. These tools can produce text that looks authoritative but is entirely fabricated. Courts expect parties, whether represented or not, to provide accurate legal references and to correct mistakes promptly. If you are representing yourself, it is essential to rely on authoritative sources such as legislation.gov.uk, official court guides and reputable legal textbooks. AI can help you understand concepts or draft documents, but it cannot replace the quality of service that can be provided by a legal professional.

Conclusion

Ultimately, Cork v Smith reinforces a simple but crucial message: whether you are instructing a solicitor or navigating the system alone, accuracy and honesty are non‑negotiable. The courts rely on them and your case depends on them. If you are unsure about the legal position or the reliability of information you have found, seeking professional advice is always the safest course.