New Zealand Law Society - Case Law AI
This selection of cases, prepared by our Library team, provides insights about how other jurisdictions are considering the obligations and responsibilities of lawyers when using material generated with AI. New Zealand cases will be added as these become available.
Our Library team provides fee-based services for research including case law, and AI citation checking. Find out more about these services, make an online request now, or contact the team at library@lawsociety.org.nz.
Please refer to Law Society guidance about using Generative AI and the Courts of New Zealand guidance about using Generative AI when representing clients in court and tribunal proceedings – available at our Artificial Intelligence Hub.
Guideline case, citation of fictitious authorities, duty to court, training, supervision
Citation of fictitious authorities, solicitors' powers and duties, negligence, wasted costs orders
R. (on the application of Ayinde) v Haringey LBC [2025] 1 W.L.R. 5147
In two cases referred under the Hamid jurisdiction, the court considered lawyers’ actual or suspected use of generative AI to produce documents containing fictitious legal authorities.
In Ayinde, a junior barrister acting in a housing judicial review cited five non-existent cases. The housing authority’s solicitor raised the issue and threatened a wasted costs application. The barrister denied using AI, but accepted that she had acted negligently and unreasonably. The court found the conduct improper and unreasonable, ordered wasted costs and referred her to the Bar Standards Board. The law centre solicitor who approved her response was also referred to the SRA because his supervision and response to the problem were inadequate.
In Al-Haroun, 18 of 45 authorities cited in witness statements were fictitious or materially misquoted. The client accepted that AI tools and online sources had generated the inaccurate material. The solicitor admitted relying on the client’s research without independently checking it and self-reported to the SRA.
Freely available generative AI tools were not reliable for legal research. Lawyers using such tools had a professional duty to verify their output against authoritative sources, including official legislation and judgment databases and reputable law reports and legal publishers. Asking AI to identify uncertainty or weaknesses did not remove this obligation. The critical safeguard was independent verification.
In Ayinde, the threshold for contempt was met but proceedings were considered inappropriate partly because of unresolved factual issues, her extreme junior status, and concerns about her training and supervision. She was referred to the Bar Standards Board. The solicitor was also referred to the SRA for inadequate action after being alerted to the problem; the paralegal was found blameless. In Al-Haroun, the solicitor’s failure to verify authorities was serious, but the contempt threshold was not met. He was referred to the SRA. The court emphasised that lawyers could not rely on clients to verify legal authorities, accuracy remained the lawyer’s professional responsibility. Lawyers could not delegate responsibility for verifying authorities to a lay client.
The case also includes an Appendix discussing similar cases and penalties from other jurisdictions.
The Upper Tribunal confirmed that the guidance in R (Ayinde) v Haringey LBC applied to solicitors and also Immigration Advice Authority (IAA)-regulated advisers using AI for legal research or drafting.
The cases involved non-existent case authorities being cited in immigration proceedings. One used Google’s AI search and ChatGPT for drafting emails and summarising Home Office decisions. In another case, false authorities were cited in work prepared by a junior trainee under supervision.
The risks of inaccurate AI-generated material extend beyond ChatGPT to AI-assisted search tools generally. While properly trained professionals may benefit from legal AI, practitioners must verify all AI-generated information against reliable legal sources before using it. The responsibility ultimately rested with the legal professional handling the matter. Supervising lawyers must ensure that junior staff’s work was properly checked and documents submitted to a tribunal contain only accurate authorities. A supervisor who failed to carry out adequate checks may be more culpable than the junior lawyer who made the error.
Uploading client letters or Home Office decisions to an open-source AI tool may put confidential information into the public domain, potentially breaching client confidentiality and waiving legal professional privilege. Closed-source AI systems may reduce these risks, but professionals should consider regulatory and data-protection requirements.
The Tribunal also introduced stronger requirements for statements of truth, requiring legal representatives to confirm that cited authorities exist, can be located using the citation given, and support the legal proposition for which they are cited. Signing such a statement where false authorities are included may ordinarily result in regulatory referral.
Ndaryiyumvire v Birmingham City University County Court (Birmingham) [2025] 10 WLUK 719 14 Oct 2025
The court made a wasted costs order against a solicitor’s firm after an application filed on behalf of a claimant contained fictitious case authorities generated by AI. The claimant had made an application to amend her particulars of claim which had referred to fictitious authorities and contained a statement of truth, beneath which the claimant's solicitor’s name (N) was printed as legal representative for the claimant. The defendant's solicitors asked N for copies of those authorities. N did not reply, instead filing a renewed application to amend asking the court to disregard the first application The firm withdrew the document, apologised and took steps to prevent a recurrence.
Lawyers who used AI to conduct legal research had a professional duty to check its accuracy by reference to authoritative sources before using it. N had put forward fictitious authorities which was improper, unreasonable and negligent conduct and had failed to adequately provide an explanation. Under the Code of Conduct for solicitors, a solicitor was accountable for work done by staff. On the question of wasted costs, the crucial issue was hot to apply Ayinde, which stated “In principle, and subject to any explanation… placing false material before the court with the intention of the court treating it as genuine amounts to improper, unreasonable and negligent conduct." Failures in administration meant that it was appropriate to make a wasted costs order. A wasted costs order was compensatory not penal.
The failure was in substance a failure of management at the firm more than N's failure as an individual solicitor. It was not appropriate to make a further specific referral of N to the SRA in addition to or separately to the wasted costs order.
Pasuengos v Minister for Immigration and Citizenship (No 2) [2026] FEDCFAMC2G 96
The applicant sought judicial review of a decision of the Administrative Appeals Tribunal. The application was unsuccessful in Pasuengos v Minister for Immigration & Citizenship [2025] FedCFamC2G 2129 (Pasuengos (No 1)). This judgment concerns the conduct of the applicant’s legal representatives after they filed submissions containing three non-existent, or “hallucinated”, authorities generated through reliance on an AI-assisted Google search.
The Minister identified the fictitious authorities before the final hearing. Counsel for the applicant confirmed the authorities did not exist and that the citations had not been independently verified. The junior solicitor accepted responsibility, explaining that the citations resulted from online research and AI-generated material that had not been checked against authoritative legal databases such as AustLII, LexisNexis, or Westlaw. The principal solicitor also accepted responsibility for failing to verify the authorities. Both apologised and outlined measures to prevent recurrence, including improved verification procedures and further AI-related training.
Using generative AI for legal research was not inherently impermissible. However, AI was not a substitute for legal research, and practitioners remained responsible for ensuring that authorities relied upon were genuine, current, relevant and actually supported the propositions advanced. The main failure was not just a failure to verify citations, but a failure to review and assess the authorities produced by the research process. Placing false authorities before the Court constituted improper conduct and breached a lawyer’s duty to the Court. Walker JA in Kaur v RMIT [2024] VSCA 264 stated that a practitioner did not discharge their duty to the Court by only ensuring the accuracy of the citation of authorities generated by AI, they must also confirm their relevance. The Court was concerned at the apparent lack of awareness of the requisite level of diligence required. Ayinde was illustrative of the duty of diligence upon those who supervised junior practitioners or who were otherwise ultimately responsible for the matter.
Having regard to all of the circumstances, it was appropriate to refer both solicitors to the Legal Profession Conduct Commissioner (SA).
Ko v. Li, 2025 ONSC 2965 (Ont. S.C.J.)
The court considered a show cause order for contempt of court against a senior counsel who filed a factum containing AI-generated fictitious case authorities and relied on two of those cases in open court. The factum had been prepared with assistance from ChatGPT, and counsel had failed to verify the authorities before filing it. Counsel, as an officer of the court, had a duty to ensure authorities cited in submissions actually existed and supported the propositions for which they were relied upon. That duty was breached when she signed, filed and used the factum without verifying its contents. It did not matter if the factum was drafted by AI or a clerk or a student. Barristers should not sign, serve, and file with a court a submission of law without first being satisfied the authorities relied upon existed and supported the arguments made.
The publicity surrounding the case had served both to publicly denounce inappropriate conduct and as general deterrence to the bar and others who might rely on AI for legal submissions. The client was also protected from being billed for inapt services. In similar American cases, courts had ordered monetary sanctions in the range of US$5,000. But US courts had different roles than courts in Canada regarding the regulation of the lawyers who appeared before them.
Counsel had accepted full responsibility and filed a corrected version containing verified authorities. She also implemented new procedures requiring independent verification of AI-generated material and undertook to complete continuing professional development on legal ethics and the risks of AI in legal practice. A further contempt hearing would consume significant court and counsel time while providing little additional public benefit. The show cause proceeding on condition that counsel fulfil her commitment to take Continuing Professional Development course.