Ruling reminds parties to check accuracy of generative chatbot output
Following the Family Court’s rejection of a request for a safety hearing, New Zealand’s Supreme Court emphasised how misusing artificial intelligence (AI) in legal proceedings had serious implications for the administration of justice and public confidence in the justice system.
The Family Court at Whangārei – the first respondent in Jones v Family Court at Whangārei [2026] NZSC 1 – dealt with lengthy, contested proceedings regarding the care of a child.
The applicant applied for a safety hearing, a non-statutory procedure for the Family Court to screen whether it would be safe to permit a parent to access the child without supervision. The Family Court dismissed the application.
The applicant applied for a judicial review. The High Court denied the application. New Zealand’s Court of Appeal dismissed the applicant’s appeal on 16 July 2025 and his application to recall the judgment on 17 October 2025.
A “McKenzie friend” assistant was supposed to manage documents, take notes, and provide silent assistance. The appeal court allowed the nominated assistant to sit in the public gallery, where they could take notes.
Before the Supreme Court, the applicant applied for leave to appeal the appeal court’s 16 July 2025 and 17 October 2025 judgments. He also applied for an extension of time, as he had not timely pursued leave to appeal the primary judgment.
The applicant challenged the appeal court’s refusal to allow him to have a McKenzie friend at the bar table during the hearing.
The Supreme Court of New Zealand granted the application for an extension of time and dismissed the application for leave to appeal both judgments. It ordered the applicant to pay $500 in costs to the second respondent, who had filed submissions.
In granting an extension, the Supreme Court noted the modest delay in seeking permission to appeal the primary judgment. In dismissing the leave application, it did not deem it necessary in the interests of justice to hear and determine the proposed appeal.
The Supreme Court saw no error in the appeal court’s consideration of the fundamental issue, which was whether the Family Court should have addressed the safety hearing application before considering the second respondent’s protection order application.
Regarding the McKenzie friend issue, the Supreme Court accepted that a court’s refusal to allow a lay litigant to have a McKenzie friend could amount to a question of general or public importance.
However, the Supreme Court saw no question of general or public importance requiring a review. Given the substantive appeal’s legal merits, it determined that the assistant’s closer presence could not have altered the outcome.
The Supreme Court saw no risk of miscarriage of justice in the civil context.
The Supreme Court noted that the applicant’s submissions cited numerous authorities that an AI application had apparently hallucinated.
The Supreme Court stressed that those filing submissions should correctly cite genuine authorities. The Supreme Court added that non-lawyers appearing in proceedings should verify the accuracy of information provided by a generative AI chatbot.
The Supreme Court emphasised that relying on false citations, including unverified output of AI applications, could constitute obstruction of justice or contempt of court in serious cases.
