Taiwo v HomeLets of Bath Ltd & Ors [2025] EWHC 3173 (KB) (03 December 2025)

The use of AI Large Language Models in litigation continues to generate headlines (and consternation from the judiciary). Indeed, in 2025, it seemed that rarely a month went by without anew case on fake AI-generated case law. December was no exception, and the High Court has now issued a further warning regarding the use of AI by litigants.

Background

The case of Taiwo v HomeLets of Bath Ltd& Ors concerned a claim for alleged harassment and assault. The claim arose from events involving the Claimant’s tenancy and subsequent interactions with the Defendants, which the Claimant asserted amounted to both physical and non-physical harassment.

The Defendants denied the allegations, and the matter proceeded to a quantum trial where the credibility of both parties was in issue. Ultimately, the court dismissed the claim, finding the Claimant to be fundamentally dishonest in relation to the allegations made. The reasons for that finding are not relevant for present purposes.

The Claimant appealed on the grounds that the judge erred in concluding fundamental dishonesty and sought to introduce fresh evidence. She also argued that the termination of her litigation friend was incorrect.

False authorities

The Claimant submitted a skeleton argument, which included reference to Irani v Dutchy Farm Kennels [2020] EWCACiv 405. In the lead-up to the hearing, the Claimant’s litigation friend maintained that he had not been involved in preparing the skeleton argument, but was searching for the authority and would provide it if he could find it.

The Court was unimpressed; a copy of Irani v Dutchy Farm Kennels could not be found because that case was not real. It also transpired that this was not the first instance of a fake authority in the Claimant’s case: earlier in proceedings she had referred to Chapman v Tameside Hospital NHS Foundation Trust [2018] EWCA Civ 2085. As the judge explained: “There is a 2016 case in the Bolton County Court between the two named parties, but there was no appeal in 2018 to the Court of Appeal and [2018] EWCA Civ 2085 is a false reference.”

Outcome

The Claimant was represented by a litigation friend, who was not himself legally qualified. In addressing the issue, the Judge held:

The reliance upon false citations is just as unsatisfactory when presented to the Court by a litigant in person (or Litigation Friend), although of course the sanction for having done so may not necessarily be the same as those applicable if a registered lawyer is responsible for the submission.

The consequences for a lawyer in these circumstances are that they are likely to be hauled up before the irregulator for misconduct. While unregulated litigation friends do not have the same regulatory concerns, any person putting false material before the Court (party, lawyer or litigation friend) runs the risk of being hauled in front of the Court for contempt (a criminal offence). Avid readers of LP Insight articles will recall that the possibility of contempt proceedings was something discussed in our previous article (where the Claimants “narrowly and somewhat reluctantly” avoided that eventuality on account of broad public interest concerns).

What is clear is that the Court will punish those who submit false cases (AI-generated or otherwise) with costs sanctions. The Judge here followed suit, noting that “the citation of false authorities [will] have …undoubtedly added to the burden on the Court and the Respondent”, before ordering the Claimant to pay 75% of the reasonable costs of the Defendants.[1]

As we move into 2026, the risks of getting it wrong when it comes to AI are still very real. The Court requires those who appear before it (whether or not qualified and/or regulated)to take ownership of their submissions and ensure that (where AI is used) the content is accurate. The consequences for failure to do so (whether financial, regulatory or criminal) remain severe.

‍

[1]On first view this appeared, to me, quite lenient. The Claimant (quite apart from the issue as to the false cases) was wholly unsuccessful and some of her grounds of appeal were “not just not reasonably arguable, but hopeless”. However, this issue came to light at a hearing of an application for permission to appeal, where a respondent’s input is optional and it will therefore usually not be entitled to any of its costs. The Judge’s order was, therefore, a fairly significant departure from the default position.