UK tax tribunal finds appellant relied on AI-fabricated case law

The UK’s First-tier Tribunal (Tax Chamber) dismissed a capital gains tax penalty appeal in which the appellant, Felicity Harber, had submitted nine previous tribunal decisions supposedly supporting her case — none of which existed. Judge Anne Redston found the citations had been fabricated by an AI tool such as ChatGPT, most likely by a friend who had assisted Harber, a self-represented litigant, with her paperwork.

The fake decisions carried plausible names, dates and summaries, but contained tells: identical phrasing across supposedly unrelated cases and, in six of the nine, American spellings such as “favor”. None could be found on the tribunal’s own website or any legal database. Harber told the tribunal she had not known the cases were invented and did not know how to check tribunal decisions herself; Redston accepted this and did not treat the submission as an attempt to mislead the court. The appeal was dismissed regardless — the fabricated cases had not been the deciding factor, and Redston said her ruling on the underlying penalty would have been the same without them.

What the judgment is cited for since is not the outcome but Redston’s comments on cost: citing invented judgments, she wrote, “causes the tribunal and HMRC to waste time and public money”, diverting resources from other court users waiting for their appeals to be heard. The case predates the more widely reported US sanctions against represented parties — Mata v. Avianca six months earlier involved lawyers, not a litigant in person — and became an early UK reference point for guidance on AI use by unrepresented parties, who make up a large share of tribunal appellants and have no professional duty of verification to enforce against them.