The professional duty to verify legal citations, and what follows from relying on generative AI output. R (Ayinde) v London Borough of Haringey.
Practice; AI and citation of authorities. The Divisional Court held that freely available generative AI tools are fundamentally unreliable for legal research as they can produce plausible but entirely fabricated citations ('hallucinations'). Lawyers bear a professional duty to verify all AI-generated output through authoritative sources (BAILII, National Archives, official law reports) before use. Those in leadership positions must implement practical measures ensuring compliance with AI-related professional obligations.
Cited by
References found in the judgments’ text, with the number of paragraphs in which each case is mentioned. Not a classification: how a case was treated is recorded under Later history.
Practice; AI and citation of authorities. Applying the Divisional Court guidance in Ayinde, HHJ Charman found that a solicitor's firm had submitted two fictitious AI-generated case citations. The firm's administrative staff had used a legal software's built-in AI research feature without verification. The judge held the conduct was improper, unreasonable and negligent, ordered wasted costs against the firm, and struck out the claim with indemnity costs. [Summary not yet checked against the judgment.]
Practice; generative AI; skeleton arguments; litigants in person. At an inter partes preliminary hearing the appellant, a litigant in person, filed a 300-page, 132,000-word skeleton argument generated with ChatGPT and invited the respondents to identify its inaccuracies by 4pm the next day. Marcus Pilgerstorfer KC held that there is no principled reason why a litigant in person should not use generative AI responsibly, but that every litigant must ensure, at a minimum, that a document submitted (i) complies with the Practice Direction, (ii) has been checked for accuracy as thoroughly as the litigant is reasonably able, so that the tribunal is not misled, and (iii) contains only relevant points focused on the best arguments; submitting AI output unchecked, or placing the onus to check on the opponent, is not acceptable, and enquiries and sanctions are likely. The EAT's powers were catalogued: rejection of a non-compliant skeleton (PD 11.6.6(a)), sanctions under PD 11.6.7 and rule 26, reduced oral time, an order that applications be made only on the Annex 2 form, certification of an application as totally without merit (Mighton), and limits on correspondence (J v K; PD 1.9.2). A single ground (fair notice of the strike-out) proceeds to a full hearing; the application to resile from the position taken at the preliminary hearing was refused (Nowicka-Price). Ayinde applied to unrepresented litigants. [Summary not yet checked against the judgment.]
“Simply submitting the product of generative AI to a court or tribunal, or placing the onus to check a document onto an opponent, is not acceptable. Where concerns arise that necessary checks have not been undertaken, or that they have been conducted inadequately, judicial enquiries and potential sanctions are likely to ensue.”Marcus Pilgerstorfer KC, at [33]
Cites
References found in the judgments’ text, with the number of paragraphs in which each case is mentioned. Not a classification: how a case was treated is recorded under Later history.