A senior insolvency judge has issued a pointed warning that lawyers cannot delegate legal research or reasoning to artificial-intelligence systems, after inaccurate statutory references generated with an AI tool reached the court twice during a routine insolvency application. The matter, involving work prepared at Pinsent Masons, has become the latest illustration of how generative AI is testing established norms of professional supervision and verification within the legal sector.
What Happened in Court
According to the account of proceedings, a junior lawyer used an AI tool while preparing material for an otherwise straightforward insolvency application. The tool produced statutory wording that was inaccurate, and the errors were only surfaced when ICC Judge Mark Mullen questioned the phrasing during the hearing. The flawed material is understood to have been placed before the court on two occasions within the same process before the discrepancies were identified.
Notably, the AI system had itself flagged that it was "not fully confident" in the statutory wording it produced, advising that the output be independently checked before submission. That caution was not acted upon. Supervising lawyers reportedly told the court they had been unaware that AI had been used at all in the preparation of the material.
The Judge's Warning
Delivering his observations, Judge Mullen stressed that legal professionals "cannot outsource legal research or legal reasoning to AI systems." The central concern he identified was one of reliance: courts depend on solicitors and barristers to verify the accuracy of legal authorities and procedural rules before they are cited. A judge, the court indicated, should not be placed in the position of independently re-checking statutory wording submitted by a major commercial law firm.
The episode was framed less as an isolated technological slip than as a question of governance and oversight, echoing a broader judicial tendency to treat AI-related failures as matters of supervision rather than one-off human error.
Regulatory Fallout
The firm apologised and made a self-referral to the Solicitors Regulation Authority (SRA). In light of that apology, the self-reporting, and an acceptance of the costs incurred by the client, the court declined to initiate contempt proceedings. The outcome of the SRA's review remains pending at the time of reporting.
Observers expect the case to sharpen regulatory attention on how firms govern AI use, including internal review procedures, disclosure of AI assistance, and the supervision of junior staff who may reach for such tools during routine work.
A Wider Pattern
The incident sits within an emerging international trend. Comparable difficulties have surfaced in the United States, including reported errors in a bankruptcy filing associated with work at Sullivan & Cromwell. Across jurisdictions, courts appear increasingly willing to characterise AI-related lapses as systemic failures of process and control, rather than as understandable individual mistakes.
The wider implications for the profession are considerable. As generative tools become embedded in day-to-day legal practice, the case underscores that responsibility for accuracy remains squarely with the humans who sign and submit the work. Firms will likely face pressure to formalise AI-use policies, mandate verification of any AI-assisted output, and ensure supervisory chains are equipped to ask whether such tools were involved. For a sector built on trust in the reliability of what it puts before a court, the message is that efficiency gains cannot come at the expense of professional judgement.
