The Supreme Court has weighed in on the use of artificial intelligence by lay litigants.
In Jones v Family Court at Whangarei,1 Mr Jones sought leave to appeal to the Supreme Court, following unsuccessful attempts to judicially review a Family Court decision declining him unsupervised access to his child.
The Court was concerned that Mr Jones had used artificial intelligence (AI) in preparing his submissions for the leave application. A number of the authorities Mr Jones cited were “hallucinated” – a phenomenon where an AI model produces nonsensical or false outputs that it presents as accurate – combining real case names with incorrect citations. The Court said:
Misuse of AI in legal proceedings has serious implications for the administration of justice and public confidence in the justice system. Persons filing submissions in court must ensure all authorities referred to are genuine and correctly cited.
The Court stated that, in serious cases, reliance on unverified hallucinated citations may have criminal consequences, amounting to obstruction of justice or contempt of court.
The rebuke of a lay litigant’s use of AI by our highest court is significant and is the strongest judicial statement yet on this issue.
Lay litigants are required to comply with Generative AI Guidelines
The Supreme Court relied on the Guidelines for Use of Generative Artificial Intelligence in Courts and Tribunals: Non-Lawyers, a set of public guidelines issued by Courts of New Zealand for the use of artificial intelligence by non-lawyers (including lay litigants). These Guidelines apply to courts and tribunals at all levels.
The Guidelines do not prohibit the use of the AI by non-lawyers. Rather, they set out some useful rules for non-lawyers to abide by when using AI in the course of proceedings, such as:
- Accuracy: AI models can often “hallucinate” fake cases, legislation, or other information. They can cite authorities which are fictitious or, if they do exist, do not contain the passages quoted from them or do not support the propositions for which they are cited.2 The Guidelines confirm that lay litigants are responsible for ensuring all information provided to a Court is accurate. The Courts have previously warned lay litigants of the dangers of relying on hallucinated case law produced by generative AI (GenAI).3
- Confidentiality: AI models such as ChatGPT ‘learn’ from the information put into them by users. The Guidelines warn that only information in the public domain should be inputted into an AI model. Otherwise, such use of AI may amount to a breach of a suppression order, statutory prohibitions on publication, or impermissible disclosure of confidential, privileged or private information. This situation previously arose in QTR v BXD, where an employee’s inputting of company intellectual property into an AI model when preparing responses in the Authority investigation process was held to be in breach of the confidentiality provisions of their employment contract.4
- Disclosure of AI use: Lay litigants are not required to disclose the use of AI, unless expressly asked or required to do so by the Court. We have seen an example of a lay litigant confirming to the ERA that they have used AI in preparing their case.5
There are parallel guidelines for lawyers, which say the use of AI in preparing submissions and other Court documents may amount to a breach of the Lawyers Conduct and Client Care Rules 2008, especially the obligation not to mislead the Court (r 13.1).6
The use of GenAI by lay litigants is increasing
We are seeing a rapid rise in lay litigants’ reliance on GenAI to prepare and present their claims. This includes claims against insurers seeking indemnity under insurance policies or against insureds who have the benefit of liability insurance policies.
There is an obvious access to justice benefit in lay litigants’ use of GenAI. They may be assisted in preparing claims in circumstances where legal fees may otherwise be prohibitive.
However, while it can make a lay litigant’s documents more comprehensible than their own words, their use of GenAI can exacerbate the difficulties faced by an insurer in dealing with unrepresented parties:
- No independent advice: A lay litigant’s claim is often brought relying on AI without the benefit of independent legal advice. The lay litigant may not understand the information generated by the AI or the limitations of GenAI in predicting the merits of their claim. This can (and does) lead to meritless claims, and irrecoverable costs incurred in defending them. The Guidelines expressly state that AI cannot be relied on to predict the chance of success or outcome of a case.
- Dense or illogical: Pleadings and submissions produced by GenAI are often still illogical and error-ridden. Defendants must incur the cost of addressing these.
Practical Guidance
Where it appears a party may be relying on GenAI, the first step is to identify if the documents produced bear the hallmarks of GenAI. Courts have previously looked at factors such as reliance on hallucinated cases, the repeated use of particular punctuation marks such as “M dashes”,7 Americanised spellings8 or submissions that are prolix and difficult to understand9 which demonstrate that a party’s case was likely prepared by GenAI.
Where it is clear that a lay litigant is using AI in presenting their claim, there are a number of practical points to consider:
- Confidentiality concerns: Does the content produced by AI rely on or refer to material that may be confidential, privileged, suppressed or otherwise private? If so, it may be necessary to seek an interim injunction or enforcement orders to protect that information.
- Hallucination: Are the citations relied on hallucinated? If they are, this can be pointed out to the Court. In extreme cases, it may amount to contempt of court or obstruction of justice.
- Costs consequences: While not addressed in Jones or the Guidelines, the use of GenAI in presenting a claim may have costs consequences for an unsuccessful party. For example, where a plaintiff has pursued a meritless argument in reliance on GenAI or has relied on hallucinated cases, an increased costs award may be sought on the basis that the plaintiff’s conduct of the case has unnecessarily contributed to the time or expense of the proceeding.10
Conclusion
We expect to see a continued uptake in the use of GenAI by lay litigants as AI models become more sophisticated, in particular to prepare and present their claims. The Supreme Court’s comments in Jones are a useful indication that the Courts are alive to the potential risks that this technology presents, and will be prepared to act where its misuse impacts on the efficacy of our justice system.
If you would like to know more about the issues discussed in this article, please contact Zane Fookes
- [2026] NZSC 1.
- See R (Ayinde) v Haringey London Borough Council [2025] EWHC 1383 (Admin), [2025] 1 WLR 5147 at [6] and [73].
- Wikeley v Kea Investments [2024] NZCA 609, [2024] 3 NZLR 901 at [199], n 187; Cunningham v healthAlliance NZ Ltd [2025] NZEmpC 191 at [36]; LMN v STC [2025] NZEmpC 46 at [8].
- QTR v BXD [2025] NZERA 716 at [21].
- O’Driscoll v Rehab Co Mobile Ltd [2025] NZERA 746 at [6].
- See also, by way of example, R (Ayinde) v Haringey London Borough Council [2025] EWHC 1383 (Admin), [2025] 1 WLR 5147, in which the UK High Court referred barristers and solicitors who relied on hallucinated authorities to the regulator.
- QTR v BXD [2025] NZERA 716 at [18(e)]. A pair of M dashes – so-named for being the width of the letter ‘m’ – are used to introduce a thought or idea that explains or expands upon that which preceded it. An overuse of M dashes is often seen as an indicator of AI-generated text (particularly ChatGPT), owing to the material on which the AI systems were trained.
- R (Ayinde) v Haringey London Borough Council [2025] EWHC 1383 (Admin), [2025] 1 WLR 5147 at [38].
- Coulson v ASB Bank Ltd [2025] NZHC 3424 at [11].
- High Court Rules 2016, r 14.6(3)(b). See, for example, O’Driscoll v Rahab Co Mobile Ltd [2026] NZERA 106, in which the ERA awarded a costs uplift in part due to the unsuccessful self-litigant referring to non-existent cases in his submissions: at [10].
This publication is intended as a general overview and discussion of the content dealt with. It should not be used in any specific situation, in which case you should seek specific legal advice.
