The Dekagram: 12th October 2026

This week Ben Rodgers considers whether forensic accountancy evidence is actually expert evidence at all; and Charlie Stonehill provides us with another warning on the use of artificial intelligence. These cases are coming thick and fast now – we respectfully suggest our readers consider how and when they use these tools, both as firms and as individuals.

White-labelling accountants’ reports

It is not unusual in injury litigation to seek from the court, and obtain, permission to rely upon forensic accountancy expert evidence at trial.  A judgment in the Chancery Division last December calls this practice into question.  We may have been getting this all wrong: we may not need permission under CPR Part 35, to introduce the work of forensic accountants into civil trials.  At least, not the type of forensic accountancy reports we get in injury cases.

The judgment is that of HHJ Paul Matthews in Mashal v. Javed [2025] EWHC 3195 (Ch).  This was a claim for repayment of £1m received by a trustee.  The Claimants had obtained permission to rely on “expert evidence in the field of forensic accounting, to trace the Investment Monies.”  The resulting report was 40 pages long.  It was well-presented.  It considered the documentary evidence available to the Claimants, including records of bank accounts.  It synthesised the information contained in those documents to summarise the available evidence.

In the judgment of HHJ Matthews, the report was a classic exercise in forensic accountancy.  But it was not expert evidence at all.  It was, at best, hearsay evidence of fact.  It was hearsay because the underlying documents themselves were the primary evidence.  The accountant was simply reporting some of what those documents said.  He was not adding any expert opinion of his own in doing so.  He was simply making information contained in a great many existing documents more digestible for the parties and the judge by reducing them to a tabular or summary format.

The accountant, as a qualified and experienced accountant, was certainly an “expert” in relation to the preparation of accounts and conformity with relevant standards.  But his forensic accountancy report did not contain opinion evidence about that.  It would have been different had he been commenting on or criticising the accounting statements he had been given.  But he wasn’t.  He was just summarising them.

In the judgment of HHJ Matthews, the evidence of a forensic accountant does not need permission under Part 35.  (Unless it contains true opinion evidence.)  It may need a hearsay notice.  Or it could simply be presented as part of the submissions of the party who commissioned it.

HHJ Matthews gave this judgment at the pre-trial review.  When the case was tried ([2026] EWHC 1269 (Ch)), the deputy High Court judge agreed that the report was not an expert report.  He nevertheless relied on it as though it was witness evidence.  He accepted all the accountant’s calculations.

About the Author

Ben Rodgers was called in 2007 and now specialises in personal injury work with an emphasis on accidents abroad, including maritime accidents (he is himself an excellent sailor). He is listed for personal injury in the Legal 500, where he is said to be ‘go-to counsel for complex liability disputes; calm and composed, but will fight ferociously when required.’

When the Tool Is Not the Problem

In NK v The Secretary of State for the Home Department [2026] UKAITUR UI2025002601, the Upper Tribunal (Immigration and Asylum Chamber), dealt with an asylum appeal that it described as having “an unhappy history, thanks to the use by the appellant’s solicitors of AI to generate the grounds of appeal and skeleton argument” [2]. Two individuals have since reported themselves to the Solicitors Regulation Authority, so the Tribunal said nothing more about professional conduct [2]. Instead, the case shows the considerable difficulties which can arise when inappropriate use is made of AI tools; it is a useful warning for all practitioners when navigating the inevitable implementation of AI into their practices. 

The Issue

The appellant, a Namibian national, claimed asylum on the basis that her family would persecute her because of her bisexuality. The First-tier Tribunal dismissed her appeal, prompting her solicitors to draft grounds of appeal to the Upper Tribunal. That is where things began to go wrong.

The grounds of appeal were, as the appellant’s own counsel later accepted, “completely infected by the use of AI” [7]: they cited fake cases, relied on paragraphs that do not appear in the judgments cited, and quoted passages from the FTT decision that the judge never wrote. One of those quotations, said to come from the FTT decision, was described by the Tribunal as “pure invention”. Even so, part of ground (3) was built on criticising it [8]–[9].

Remarkably, these problems had already been flagged. When the FTT judge granted permission to appeal, he noted that the grounds misquoted the decision in numerous places [12]. The solicitors did not take the hint. Rather than amending the grounds, they filed a skeleton argument in the Upper Tribunal based on the same defective material [13]. The problem came to a head only when the Home Office Presenting Officer asked for clarification of the citation errors the day before the hearing. At that point the solicitors admitted that AI had been used to generate both the grounds and the skeleton argument, and the hearing had to be adjourned [13]– [14].

The cost to everyone involved was considerable. The Tribunal noted that the issues which “occupied a great deal of time” at the resumed hearing arose only because of Sarker Solicitors’ “inappropriate use of AI” [18]. The appellant herself paid the heaviest price. The Tribunal refused permission to add an entirely new ground and a reformulated ground, reasoning that allowing them would let a party gain an advantage by filing false or misleading grounds in time and then replacing them once the falsehood was discovered [65]. The one surviving ground disclosed no error of law, and the appeal failed. The Tribunal has also given directions for the Home Office to pursue a wasted costs application against the firm [76].

Key Learning Points

First, and no doubt a point that will be emphasised repeatedly as we adapt to AI, the responsibility rests with the lawyer, not with the tool. The most important passage in the decision is at [44], where the Tribunal said it would be misleading to describe the irregularities as caused by AI “rather than by the solicitors’ misuse of AI” [44]. The grounds misrepresented what the judge had said, what the evidence showed and what the authorities decided, all because the firm did not check the grounds, did not consider the grant of permission and did not check the skeleton argument. The lesson is simple but essential: AI output is at most a first draft. Every citation, quotation and statement about the evidence must be checked against the original source before it is put before a court or tribunal, because the person who files the document owns its contents.

Second, supervision and confidentiality cannot be an afterthought. Both documents were produced by a paralegal using ChatGPT [44]. The Tribunal observed that this “itself raises significant concerns about client confidentiality and legal privilege” [44]. Firms and chambers should therefore have clear policies on which tools may be used, what client information may be entered into them, and how AI-assisted work by junior colleagues is reviewed.

Third, (and applicable to legal practice as a whole, not just when using AI), when errors come to light, deal with them promptly and candidly, because the client bears the consequences. The Tribunal stressed that a representative who discovers a misrepresentation in their case must identify it “precisely and promptly” [48]. In this case, the errors were eventually identified only after the Home Office raised them, with counsel finding at least 18 separate errors [48]. The Tribunal noted that the sooner such misstatements are identified, the greater the chance of avoiding an unnecessary hearing and wasted costs [49]. The failure to act on the FTT judge’s early warning was costly [12]– [13]. Above all, the appellant, whom her own counsel described as blameless, lost the chance to advance her best case.

Conclusion

This case is not a condemnation of AI, which is refreshing given the inevitability of it making its way into our personal and professional lives. However, it is a condemnation of unverified and unsupervised reliance on it. The Tribunal’s message is clear: practitioners who use AI tools must take full responsibility for what they file, and failing to do so can harm the client, the tribunal and the practitioner alike.

About the Author

Charlie Stonehill was called in 2021 and joined Deka after successfully completing pupillage in January 2026. He now undertakes work across all of chambers’ practice areas. In his spare time he enjoys participating in and attending sport, music performances, and theatrical productions; and enjoying good food.