Felicity Harber submitted nine fictitious AI-generated case citations to a UK tribunal

Felicity Harber, a litigant in person appealing an HMRC penalty for failure to notify Capital Gains Tax liability, submitted nine fabricated First-tier Tribunal case citations generated by an AI system such as ChatGPT. The Tribunal found that none of the cited cases existed on any legal database, though they bore superficial similarities to real cases. The Tribunal accepted Harber was unaware the cases were fabricated but dismissed her appeal and warned that citing invented judgments wastes public money and undermines confidence in the judicial system.

Felicity Harber · Incident Dec 4, 2023 · Indexed Jun 4, 2026 · 3 sources

An AI chatbot fabricated nine plausible but nonexistent tribunal decisions that a self-represented taxpayer submitted as legal precedent, exposing how generative models hallucinate authoritative-sounding citations with no basis in reality.
What
Felicity Harber, a litigant in person appealing an HMRC penalty for failure to notify Capital Gains Tax liability, submitted nine fabricated First-tier Tribunal case citations generated by an AI system such as ChatGPT.
Incident date
Dec 4, 2023
Who
Felicity Harber
Failure mode
Hallucination
AI surface
Chatbot
Severity
Medium

What happened

Felicity Harber appealed an HMRC penalty for failure to notify Capital Gains Tax liability after disposing of a property. As a litigant in person, she submitted a written response containing nine FTT case names, dates, and summaries she claimed came from a friend in a solicitor's office. HMRC counsel could not locate any of the cited cases on legal databases. The Tribunal found that all nine were fabricated by an AI system such as ChatGPT, with names and details bearing superficial similarity to real cases but being entirely fictitious. The Tribunal accepted Harber did not know the cases were fake but dismissed her appeal, noting the submission had wasted tribunal and HMRC time and public money.

What broke inside the model

Failure path · mode profile · Hallucination
  1. 01 · TriggerA user asks for a fact, a citation, or a figure.
  2. 02 · Model stepThe model writes a fluent, confident answer.
  3. 03 · Control gapNothing ties the claim back to a real source.
  4. 04 · FailureA fabricated fact ships as if it were verified.
  5. 05 · ConsequenceThe false claim reaches a customer, a court, or the public.

Confidence holds, and even spikes, as the claim detaches from any source.

The AI chatbot (likely ChatGPT) produced nine plausible-looking but entirely fabricated legal case citations, including names, dates, and summaries that mimicked genuine tribunal decisions. The model's statistical text prediction generated output structurally consistent with real case law but with no grounding in actual judicial records, a phenomenon where the system produces highly plausible but incorrect results. Without any verification step or access to a real legal database, the fabricated citations appeared authentic enough to be submitted as legal authority.

Public visibilityHigh
Regulatory exposurePossible
Customer impactFew customers
Financial impactUnknown
Time to disclosureDays
  1. Court FilingHarber v Commissioners for His Majesty's Revenue and Customs [2023] UKFTT 1007 (TC)bailii.org
  2. PressLitigant unwittingly put fake cases generated by AI before tribunallegalfutures.co.uk
  3. Reader-Submitted12 False Citations/AI Hallucinations in UK Courts (Ayinde)naturalandartificiallaw.com
Permalinkhttps://failureindex.ai/failures/felicity-harber-submitted-nine-fictitious-ai
CitationAI Failure Index. "Felicity Harber submitted nine fictitious AI-generated case citations to a UK tribunal" (FI-0122). Realm Labs. https://failureindex.ai/failures/felicity-harber-submitted-nine-fictitious-ai (indexed Jun 4, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0122. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would have caught this

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.