Artificial intelligence is transforming the legal and insolvency sectors at remarkable pace.
As R3's recent report confirms, law firms and insolvency practices of all sizes are increasing their use of generative AI tools, both the ubiquitous tools such as CoPilot and ChatGPT, and specialist legal platforms such as Legora, to review documents, manage large case portfolios, prepare reports, and assist with the volume of administrative work that accompanies any instructions and appointment.
Used properly, AI offers genuine benefits to the insolvency profession. Large language models can review and summarise vast quantities of documentation almost instantaneously, a task that might otherwise take days of manual review. They can produce well-structured first drafts of correspondence, chronologies, document reviews and reports, freeing practitioners and their advisers to focus on substantive analysis and decision-making.
For smaller firms, AI can level the playing field, providing access to capabilities that were previously the preserve of larger practices with greater resources. The technology also supports consistency and efficiency across routine processes, such as statutory notifications and creditor communications.
For office holders navigating complex and time-pressured appointments, as well as dealing with volume / process driven tasks, the potential efficiencies and cost savings can be very appealing. But as a recent High Court judgment starkly demonstrates, the possible gains come with significant risks that every practitioner must understand.
The risks — Cork v Smith [2026] EWHC 1199 (Ch)
A critical limitation of generative AI is that it predicts rather than understands. It can produce text that appears to be factually accurate, without knowing whether it is indeed correct. The phenomenon known as "hallucination", where AI generates false or misleading information and presents it as fact, is a well-documented and inherent risk. An AI tool may cite a specific statutory provision, quote a particular paragraph number, and present fabricated text with complete confidence. The output will be fluent, well-formatted, and entirely wrong.
This risk is not confined to complex or contested matters. The case of Cork v Smith [2026] EWHC 1199 (Ch) arose from a block transfer application, one of the most routine applications in insolvency practice. A junior associate solicitor used an AI tool to research whether the Court had power to order a liquidator's release in the context of a voluntary liquidation. The AI generated a fictitious quotation of Rule 12.37(5) of the Insolvency (England and Wales) Rules 2016, and the generated text bore no resemblance to the actual provision. It was then included in a letter to the Court without verification, despite the AI itself having expressly warned the user to check the wording against the primary legislation before citing it.
The position was compounded when the Court queried the fabricated text and the lawyers in question, rather than investigating and acknowledging the error, sought to justify the submission by characterising the fictitious quotation as a "summary conclusion" drawn from reading various provisions of Rule 12.37. The judge described himself as "astonished" by this response, and directed that the matter be listed for a hearing with witness evidence. The 59 pages of AI chat logs that were submitted in evidence confirmed the hallucination and the failure to heed the AI's own warning.
Practical tips and best practice
As the President of the King's Bench Division made clear in R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin), those who use AI to conduct legal research have a professional duty to check the accuracy of such research by reference to authoritative sources before using it in the course of their professional work. That duty rests on the lawyer who uses the AI and on any lawyer who relies on the work of others who have done so.
Insolvency practitioners, often acting in a capacity where they are officers of the court, have similar duties and responsibilities to the Court. This means that an office holder must not be indifferent to the accuracy of material presented to the Court on their behalf, whether in applications, reports, or supporting evidence.
We therefore suggest the following practical tips and best practices when using AI and reviewing AI generated documents.
- First, treat AI as a starting point, never a finished product. Every legal proposition, rule reference, and statutory quotation must be verified against the primary source before it is placed before the Court. In Cork v Smith, a two-minute check on legislation.gov.uk would have prevented the entire episode.
- Second, engage with your legal advisers about their use of AI. Ask what tools are being used, how outputs are verified, and what supervision processes are in place. Any firm with robust procedures will welcome the question.
- Third, do not assume that routine matters carry lower risk. Block transfers, statutory notifications, and standard applications which can be processed more quickly may be subject to less scrutiny, precisely the conditions in which errors can pass undetected.
- Fourth, be transparent if things go wrong. In Cork v Smith, the Judge noted that had the true explanation been given promptly with an appropriate apology, no further action might have been necessary. The attempt to justify the error proved worse than the original mistake.
- Finally, be alert to AI use by directors and other parties. Information presented to you as fact, whether in statements of affairs, correspondence, or proposals, may itself have been generated by AI and should be treated with appropriate caution.
AI is here to stay, and used properly it can deliver genuine efficiencies for practitioners and their advisers alike. But it is a tool to be used with care — not a substitute for human expertise and oversight.
