Fake AI Citations in Court: Lessons for UK Solicitors

What the Divisional Court decided in Ayinde

Two cases were referred to a Divisional Court and heard together under the Hamid jurisdiction, the court’s inherent power to regulate its own procedures and enforce the duties lawyers owe it. Dame Victoria Sharp P and Johnson J gave judgment on 6 June 2025.

In Ayinde, grounds for judicial review contained five citations that did not exist, one purporting to be a Court of Appeal decision, along with a misstatement of a statutory provision. When the defendant’s solicitor challenged them, copies could not be produced. Ritchie J, hearing the wasted costs application, called it wholly improper to put fake cases in a pleading and rejected the description of them as minor citation errors or cosmetic errors. He ordered the claimant’s solicitors and counsel each to pay £2,000, referred the barrister to the Bar Standards Board and the solicitors to the SRA.

In Al-Haroun, the solicitor accepted that he had relied on legal research carried out by his client without independently verifying the authorities. He referred himself to the SRA, and the court referred him as well. The Divisional Court concluded the threshold for contempt proceedings was not met in either case, but sent the judgment to the regulators and invited them to consider urgently what further steps to take.

The principle that came out of it is the one that matters for practice: where a lawyer cites false authorities, whether through unchecked AI use or otherwise, it will usually be appropriate for the court to refer the matter to the regulator, and to consider other sanctions.

What has happened since

The judgment did not stop the problem. The Upper Tribunal reported a considerable increase in the citation of fictitious authorities through the second half of 2025, and around fifty such cases have now been reported in England and Wales.

Two later decisions extend the picture. In Ndaryiyumvire v Birmingham City University, HHJ Charman declined to make a separate SRA referral against a solicitor, finding that administrative staff at the firm rather than the solicitor had placed the false material before the court, though a wasted costs order was made and would itself reach the regulator. And in R (Munir) v Secretary of State for the Home Department [2026] UKUT 81, the Upper Tribunal raised what appears to be a new point: that putting client letters and Home Office decision letters into an open tool such as ChatGPT can amount to placing that information in the public domain, with consequences for confidentiality and privilege.

Closer to home for solicitors, a judge referred two solicitors to the SRA in 2026 after inaccurate authorities appeared in court documents, where the fee earner accepted he had not verified research carried out by a paralegal and acknowledged AI tools may have been used. We covered that case in Judge refers solicitors to SRA over AI-generated citations. Judge Grimshaw declined contempt proceedings, finding insufficient evidence that the solicitors knowingly put false material before the court rather than acting negligently or recklessly.

The regulatory and judicial guidance

Guidance has accumulated quickly since Ayinde. The Courts and Tribunals Judiciary’s AI guidance for judicial office holders was updated in October 2025 and states that all legal representatives are responsible for the material they put before the court and have a professional obligation to ensure it is accurate and appropriate. It warns that AI tools are a poor way of finding new information that cannot be independently verified. The Bar Council issued updated generative AI guidance in November 2025, and the Civil Justice Council published an interim report and consultation on the use of AI in preparing court documents in February 2026.

For solicitors, the SRA’s position runs through its existing standards rather than a separate AI rulebook. The duties to uphold the proper administration of justice, to act with integrity, to maintain competence, and to protect client confidentiality all apply to AI-assisted work exactly as they apply to any other work. That is the practical significance of Munir: the confidentiality exposure from a public AI tool is a regulatory problem before it is ever a litigation problem.

What firms should actually change

The recurring feature of these cases is not that a lawyer used AI. It is that nobody checked, and that the checking gap sat somewhere the supervision structure did not reach: a pupil, a paralegal, a client, administrative staff. Three controls follow from that.

• Verify every authority against a reputable source before filing. Not a search result summarising the case, the judgment itself on a recognised database or on Find Case Law. A citation that cannot be opened does not go in the bundle
• Make verification an accountable step, not an assumption. In both Al-Haroun and the 2026 SRA referral, the failure was a supervising fee earner relying on research done by someone else. Whoever signs the document owns the citations in it
• Set a rule on which tools may touch client material. Munir puts the confidentiality point beyond argument for open, public tools. Firms need a stated position on what may be uploaded where, and staff need to know it before they are under deadline pressure

Training matters more than policy here. Fee earners need to understand that a fabricated citation from a language model looks exactly like a real one, which is precisely why it survives a skim read. If your firm is assessing where it stands on this, our AI Compliance Readiness Score covers verification, supervision and confidentiality among other dimensions, and sits alongside our other free tools for solicitors and law firms.