Don't take AI's word for it: What a recent High Court case means for businesses using artificial intelligence

Case

Cork v Smith [2026] EWHC 1199 (Ch), Chancery Division, ICC Judge Mullen, 22 May 2026.

Executive summary

A recent High Court judgment has provided a striking and very public illustration of what can go wrong when businesses rely on AI-generated legal content without proper expert verification. Businesses that use, or are considering using, AI for legal purposes should take careful note.

Background

Under the 2016 Insolvency Rules (the Rules) the court has power to remove and replace insolvency practitioners across all appointments simultaneously via a "block transfer application". In this case, the court asked the applicant law firm to identify the legal basis for releasing the outgoing insolvency practitioner from liability as liquidator, without recourse to the Secretary of State.

Key issues

Rather than reviewing the Rules directly, a junior solicitor used the firm's AI tool to find the answer. The AI asserted that Rule 12.37(5) provided the relevant authority. It claimed the court had broad discretion when replacing an office-holder, including the power to release them from liability. It produced what appeared to be the precise statutory wording of that provision, complete with numbered sub-paragraphs. Failing to verify the AI’s output, the firm wrote to the court citing Rule 12.37(5) and presented the AI's output as a direct quotation from the legislation.

In fact, the wording the AI had produced simply did not exist. It was a “hallucination” - a fabrication generated by the AI and presented as fact.

When the court raised its concerns, the firm again turned to AI to assist in preparing its response. Rather than acknowledging the error, the AI produced a second hallucination: it asserted that the wording had never been intended as a direct quotation at all, but was merely a "summary conclusion" drawn from reading various provisions of Rule 12.37 together.

This further explanation was again sent to the court without proper verification by the lawyers themselves. The judge found the law firm’s explanation of its original correspondence as a “summary conclusion” impossible to accept, noting that the attempt to strain the language of Rule 12.37 was not credible. The judge accordingly listed a hearing to determine how court correspondence had come to refer to a rule of insolvency law that did not exist.

Outcome

Witness statements filed by the law firm confirmed that AI had indeed been used in the production of both letters. The court concluded that there was a prima facie case of breach of the duty not to mislead the court and the duty not to waste court time, but concluded that the publication of its judgment - which served as a public admonishment of the law firm - together with referral to the regulator, was the proportionate response.

More pertinently, the court was clear about one thing: the process of legal research should not simply be outsourced to an AI.

Key takeaways

The lessons of Cork v Smith are a timely reminder of the risks of placing undue reliance on AI for legal purposes. The following points are worth reflecting on carefully:

• AI hallucinations can occur early and escalate rapidly, even when AI is asked to explain itself. The AI chat transcripts, which the court set out in full in its judgment, demonstrate that hallucinations began at an early stage and compounded as further questions were asked. Critically, when the AI was asked to explain its reasoning, it did not acknowledge the fabrication - it produced a second hallucination as justification. Asking AI to interrogate its own output is no safeguard against further error

• AI will present fabricated content with the same tone and confidence as accurate content. Unverified AI-generated content, including a purported quotation from statute that had been entirely fabricated, was put before the court. Plausible-sounding output is not the same as accurate output

• AI is a tool to be used with caution, and its use does not diminish the value of cognitive analysis or professional advice, nor the need to verify accuracy by reference to authoritative sources, before relying on any output. Where decisions or documents produced with AI assistance carry commercial, legal, or regulatory consequences, they must be reviewed by a qualified professional before being relied upon

• The regulatory and judicial environment is tightening. The court set out the AI chat transcripts in full in its judgment. Courts are now ordering disclosure of AI chat logs, asking directly whether AI was used in the production of documents, and reporting professionals to regulators. This scrutiny is not limited to courtrooms - it reflects a broader and growing expectation that AI output will be properly supervised wherever it is used in a professional context

If you have questions about any of the issues raised in this article, or require cognitive legal advice from a qualified specialist, Freeths LLP has dedicated teams including data, privacy, AI and disputes who are on hand to help. Please do not hesitate to get in touch.