England and Wales: in Baker v Bath & North East Somerset Council [2026] EWHC 2706 (Admin), the High Court made no criticism of a litigant in person over a probable AI hallucination in her skeleton argument

Key takeaways

• The appeal was dismissed on the construction of Class U of the Council Tax (Exempt Dwellings) Order 1992. The AI passage is a separate section of the judgment.
• The judge made no criticism of the appellant, who he said was frank with the court and who corrected the error.
• The judgment cites the October 2025 judicial AI guidance and the Divisional Court in Ayinde, and calls the need to check AI output against authoritative sources "worth reiterating".
• The respondent council did not attend the hearing, and the judgment does not address how an opposing lawyer should handle such an error.

What did the High Court decide, and was AI the reason?

The court dismissed a statutory appeal by Ms Baker, who appeared in person, against a Valuation Tribunal for England decision of 6 February 2026. The Tribunal had upheld the council's view that her home was not an exempt dwelling under Class U of the Council Tax (Exempt Dwellings) Order 1992 for the period from 19 November 2022. The judge held that the Tribunal made no error of law: a dwelling occupied by a severely mentally impaired person together with a carer who is not a "relevant person" is not "occupied only" by persons within either limb of Class U. The household keeps the 50% discount under section 11(2)(b) of the Local Government Finance Act 1992, which the judgment says reduces the tax by half but does not make the property exempt. The judgment is explicit that AI was not the reason. In the judge's words, "the error in the skeleton argument is not the reason the appeal fails."

What did the judge say about the appellant's use of AI?

The section headed "The use of artificial intelligence" is two short paragraphs. The first says the judge makes "no criticism of the appellant for seeking help from a generative AI tool", calls it understandable that some litigants turn to such tools, and records that she was frank with the court. It says she acknowledged and corrected the error the tool created when it inserted wrong wording into Class U, "which was a probable 'hallucination'", and thanks her for that. The second paragraph says the October 2025 judicial guidance on AI recognises that these tools can produce material that is plausible but inaccurate, and that the Divisional Court in R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin) also emphasised the need to check AI-generated material against authoritative sources. The judge calls that "worth reiterating". Our articles on the judicial AI guidance and on Ayinde cover those instruments.

How did the AI error surface in the skeleton argument?

Two passages in the judgment explain it. The skeleton argument contained hyperlinks to case law authorities, and the judge records that "Those links led to a blocked ChatGPT project", which prompted him to ask whether she had used ChatGPT. She confirmed that she had. Later, in dealing with the first ground of appeal, the judge notes that the skeleton quoted the words "either alone or together with other disregarded persons" as if they came from Class U and described the wording as "clear and complete". Those words do not appear in Class U or anywhere else in the Exempt Dwellings Order, he found, so the first ground depended on a version of the provision that does not exist. The appellant acknowledged she had spotted the error before the hearing, and apologised for it. The judgment says the use of ChatGPT did not affect how he approached the substance of her arguments, which he considered on their merits.

What does this ruling add, and what does it not add?

It adds a recorded example of a court responding to a self-represented party's AI error with understanding rather than a sanction. The judge does not make a finding of misconduct, order costs, or refer anyone anywhere, and the judgment states none. It also does not announce a new test or practice direction. The High Court was reviewing the Valuation Tribunal for error of law on a council tax appeal, a judge sitting as a judge of the High Court gave the judgment, and the AI section reiterates existing guidance rather than deciding a point of law. Two features of the facts limit how far it reaches. The appellant disclosed the tool's role when asked, corrected the error and apologised. The respondent did not attend the hearing. The judgment does not say how a court would treat concealment, repeated errors or a represented party.

What might lawyers dealing with litigants in person take from this?

The judgment records that the quoted words do not appear in Class U or elsewhere in the Exempt Dwellings Order. The steps below are the desk's suggestions for lawyers dealing with litigants in person. The judgment does not prescribe any of them.

• Suggestion: check any quotation of legislation or authority in a skeleton, yours or an unrepresented party's, against the official text before you rely on it or respond to it.
• Suggestion: if you find wording that does not exist, raise it with the court and the other side plainly and without allegation, and keep a note that you did so.
• Suggestion: if a link in a bundle or skeleton does not resolve to the authority it cites, treat that as a prompt to check the underlying citation.
• Suggestion: if your own team drafts with AI tools, keep a record of who verified each quotation. The judgment refers to checking AI output against authoritative sources but does not mention recordkeeping.

What are the key dates and parties?

What we did not verify

What we opened: the full text of the judgment as published on the National Archives Find Case Law service, on 25 October 2026.

What we did not open: the Valuation Tribunal decisions, the appellant's skeleton argument itself (we rely on the judge's description of it), the October 2025 judicial AI guidance, the Ayinde judgment, and any official report or appeal history. We did not check whether permission to appeal or any further step exists.

What we refuse to claim: that the ruling creates a new rule or practice, that courts will always respond this way to AI errors by litigants in person, or anything about how the court would treat a represented party. We also do not describe the appellant's circumstances beyond what the judgment says about the appeal.

Related briefings

• AI Regulation News hub
• Europe AI regulation tracker
• Attorneys and AI regulation
• High Court on a ChatGPT-assisted witness statement and reduced weight
• Ayinde and AI-fabricated citations
• Judicial AI guidance for office holders, October 2025
• EAT on generative AI in a skeleton argument and personal responsibility