analysisAI HallucinationsAI in Courts
Harber v HMRC: the first reported UK case of AI-invented case law
The first reported UK case of AI-invented case law: a litigant cited nine ChatGPT-fabricated decisions to the Tax Tribunal, which found them fake.
Before the United Kingdom’s First-tier Tribunal (Tax Chamber), a litigant in person defending a capital-gains penalty handed up nine tribunal decisions that supported her case. None of them existed. Harber v HMRC [2023] UKFTT 1007 (TC), released on 4 December 2023, is the first reported UK judgment to find that a party had relied on case law invented by an AI system.
What happened
Felicity Harber, unrepresented, appealed a penalty imposed for failing to notify a liability to capital gains tax. To establish a “reasonable excuse,” she produced the names, dates, and summaries of nine First-tier Tribunal decisions in which, she said, other taxpayers in similar positions had succeeded.
HMRC’s representative could not locate any of the nine. The tribunal then examined the list itself. The decisions did not appear on the tribunal’s own database or on the main legal search systems; the American spelling and the repetitive structure did not read like genuine FTT judgments; and the summarised reasoning did not match how the tribunal actually decides such cases.
What the tribunal decided
The tribunal made a finding of fact about where the cases came from:
“we find as a fact that the cases in the Response are not genuine FTT judgments but have been generated by an AI system such as ChatGPT.”
It accepted that Mrs Harber had not known the cases were fabricated and had not set out to mislead. That did not save the appeal. Drawing on the US decision in Mata v Avianca — the New York case decided six months earlier on materially the same facts — the tribunal explained that citing invented authorities is not harmless: it wastes the tribunal’s and HMRC’s time and public money, and it diverts resources from other court users. On the substance, it confirmed the penalty and dismissed the appeal.
Why it matters
Harber is the baseline of the UK record. It arrived before any of the judicial or regulatory guidance now in force, and it set the pattern that later UK cases repeat: a self-represented litigant, a general-purpose chatbot, and a set of confident citations that no one checked against a law report.
Two features carried forward. First, lack of intent to deceive mitigates but does not cure — the fabricated authorities were treated as a real problem for the court regardless of the litigant’s good faith. Second, the tribunal reached for a foreign judgment, Mata v Avianca, because there was no domestic authority yet. Within two years there would be a domestic one: the Divisional Court’s ruling in Ayinde v Haringey, which set out how English courts will handle the same conduct by lawyers rather than litigants.