New Zealand

Information uploaded as at June 2026

AT A GLANCE

New Zealand Police are increasingly using AI for crime detection and prevention, particularly in operational support, risk assessment, and data analysis. Tools assist with dispatch coordination, workflow automation, frontline decision-making (e.g. SearchX and Initial File Assessment scoring), and emergency location tracking, while AI-driven analytics support vehicle tracking (ANPR), retrospective facial recognition (not live), video analysis (e.g. BriefCam), digital forensics (e.g. Cellebrite), and child exploitation investigations. Some technologies—especially ANPR and earlier Clearview AI trials—have prompted legal and privacy challenges, with appellate review ongoing as of March 2026. There is no reported AI use by prosecutors or courts, though courts are developing a digital case management system (Te Au Reka). AI adoption by defence counsel varies, with at least one judicial decision questioning the weight of AI-assisted submissions. Training on the responsible use of AI in the New Zealand justice system is not yet uniform across all justice actors. Training remains fragmented, with a mix of mandatory tool-specific instruction, optional professional development, and ad hoc initiatives rather than a coordinated national framework.

New Zealand has no overarching AI-specific regulation for the justice system, relying instead on existing technology-neutral legal frameworks (procedure, evidence, privacy, human rights) supplemented by judiciary, Law Society, and Police guidance. Courts distinguish permissible low-risk AI uses from discouraged higher-risk ones, and disclosure is generally not required unless asked. In case law, the Supreme Court's Jones v. Family Court at Whangārei [2026] warned that reliance on AI-hallucinated citations may amount to obstruction of justice or contempt of court, while the Court of Appeal's Yorston v. Attorney-General [2026] confirmed automated government systems remain subject to judicial review despite producing legally consequential records.

USE

Law enforcement 

In April 2024, New Zealand Police identified crime detention and prevention as the two key areas that the Police hoped AI would assist with. In October 2025, New Zealand Police published a Technology Capabilities List that identifies a range of AI capabilities in trial and active use, including operational and investigative tools as well as administrative support technologies. The specific tools are discussed below. Some tools are not publicly discussed to preserve operational effectiveness.

Operational support 

New Zealand Police has partnered with Appian to enhance efficiency. Appian is a tool that streamlines workflow and enhances cooperation across New Zealand’s 12 police forces. The Police’s backlog of cases, at one point up to 5,000, has been significantly reduced, with waitlines dropping from two weeks to four hours. The technology substitutes basic administrative tasks that are repetitive in nature and has allowed police members to engage in live casework.

New Zealand Police describe Microsoft 365 Copilot Chat as being in active use. Copilot Chat can assist with public-source information retrieval, preliminary data cleaning and analysis, communication tasks such as drafting content or summarising documents, and producing training resources. It draws on external web information or documents manually provided by the user.Direct access to data from Police devices or the Police enterprise system is not permitted, and inputs from New Zealand’s Police are not used to train underlying AI models. Staff must complete mandatory training before being granted access.

Azure OpenAI Information Assistant is also being trialled by New Zealand police as at June 2026, and is intended to enable natural-language search across constrained internal sources. The tool is exposed only to Police instructions (internal policies and rules for staff). The tool is not exposed to any information about people or the public, and is not intended to affect Police interactions with the public. Information is retained within the Police’s Microsoft tenant boundary to prevent data leakage.

New Zealand Police’s Online 105 (non-emergency) reporting forms use AI to scan submitted reports for keywords and assign a priority flag based on content, including offence circumstances, keywords, and sentiment, so that particular reports can be surfaced and actioned rather than processed strictly sequentially.

Hubstream is also used for case management and referral workflows, linking information to generate insights and help officers prioritise. Its primary use is as a referral system for online child exploitation, and to collaborate with partner agencies on these investigations.

OnDuty is an operational application that automatically links frontline staff to key information, particularly for family violence events. Officers attending these events can have direct links to information in the National Intelligence Application, including necessary background detail on people involved in their history. This allows staff to continue to help those in need, without spending time on recording some of the information. The new information gathered during time spent with families is stored digitally within the app, meaning less time spent writing reports back at the station. The app also offers a new approach to gathering and eliciting information at the scene, including questions in a number of different languages.

New Zealand Police have also acknowledged the use of tools that support locating individuals or devices in emergency or safety contexts, for example mobile location tools used to locate missing persons or persons at risk. The tools offer retention controls and privacy-code governance.

Predictive analytics

Since 2010, New Zealand Police have used an Initial File Assessment score, which provides a numeric value derived from weighted factors such as degree of suspicion, suspect description or identity, and vehicle information to provide an indication of case “solvability” and to support triage of case progression. The assessor reviews a file and uses either: (1) the National Intelligence Application Initial File Assessment, the Police’s core intelligence and information database, or (2) an Appian workflow calculator to compute the score. The tool is described as a structured evaluative guide rather than an automated decision-maker. While the Appian platform includes AI and robotic process automation capabilities, Police say that they are not using those features at this time.

New Zealand Police also use two complementary algorithmic risk-screening tools for family violence cases. Though these tools do not use AI, they apply a degree of algorithmic decision-making to produce a risk score and are classified as “moderate-high” risk under the Algorithm Charter (see below).

• SAFVR (Static Assessment of Family Violence Recidivism) draws on data already held by the Police (such as previous history and convictions) and is accessed by attending officers through their phone at the scene or a computer at a police station. It produces a high, moderate, or low-risk grading.
• DYRA (Dynamic Risk Assessment) is completed at the scene based on a series of structured questions asked of the primary victim, and an overall concern rating is calculated.

Data review and analysis

New Zealand Police use a wide range of tools to collect, search, correlate and analyse data in operational investigations, including platform-based evidence management, video analytics, biometric matching, mobile device forensics and investigative case management. New Zealand Police have acknowledged use of multiple systems for investigative case management and controlled searching across repositories, including the Investigative Management Tool and Investigation Search Tool.

Police also describe the use of analytics and reporting platforms such as Business Objects and SAS Visual Analytics for extracting, transforming and visualising data and producing dashboards and official statistics.

Moreover, from April 2025, New Zealand Police have been trialling MOBILedit Forensic Express, a specialised tool designed to extract and analyse data from wearable devices, such as smartwatches, in addition to mobile phones. The tool can help classify the data or media found on the device.

Police use Automatic Number Plate Recognition tools for vehicle tracking, including for real-time alerting and historical review, drawing on fixed, mobile, and third-party camera networks accessed through provider platforms, including Auror since 2014.

Automatic Number Plate Recognition use in New Zealand has attracted controversy and been the subject of legal challenge. The controversy arising from the ability to harvest historical data through large-scale access to privately-owned camera networks, allowing police to track movement retrospectively, has given rise to questions of whether this is effectively warrantless surveillance. Concerns have also been raised over a lack of oversight. Media reporting indicates that anyone with a police email address can access Auror (approximately 8,500 people), and does not require a warrant or judicial oversight. In parallel, the legality of Automatic Number Plate Recognition-derived evidence, particularly retrospective access to private camera networks, has been tested in a number of criminal cases. In 2024, the District Court rejected a challenge that this AI use required a warrant or production order, treating the technology as facilitating access to ubiquitous public space CCTV rather than constituting an unlawful search. In 2025, the Court of Appeal heard a challenge as to whether Automated Number Plate Recognition evidence must be accompanied by a warrant or production order. As at June 2026, the Court of Appeal’s decision is pending.

In September 2024, New Zealand Police disclosed that since 2022, New Zealand Police had used facial recognition technology 89 times to identify suspects. They stipulated, however, that none of these were ‘live’ searches. As an example, Automated Biometric Identification Solution (‘IBIS2’) is used in investigations for retrospective matching of images of suspects against police databases which contain lawfully held images. BriefCam also aggregates video footage, including facial recognition and vehicle licence plates, and is used by New Zealand Police to track a person or vehicle of interest. Police claim BriefCam reduces the time to analyse three months of CCTV footage from six weeks to two hours.

In 2020, New Zealand Police conducted a trial of Clearview AI facial recognition software without consulting Police executive leadership or the Privacy Commissioner, prompting concern from both the Police Commissioner and the Privacy Commissioner once the trial became public. The Independent Police Conduct Authority later oversaw a review that found that while a Police internal governance group approved the Clearview trial, it was undertaken without Police Executive approval and without consultation with the Office of the Privacy Commissioner, and it recommended stronger central governance for emergent technologies. In December 2021, following a joint Inquiry by the Office of the Privacy Commissioner and the Independent Police Conduct Authority, the Office of the Privacy Commissioner issued a compliance notice requiring Police to stop unlawfully collecting photographs and biometric prints (particularly of young people) and to delete unlawfully collected material. 

New Zealand Police have acknowledged using Cellebrite, a controversial phone hacker technology which extracts personal data from devices and can access more than 50 social media platforms, including Instagram and Facebook.

In cases relating to drug harms, police use the Lumi Drug Scan tool, a frontline capability that allows officers to test suspected drug samples in the field in real-time. The tool uses AI in the form of cloud-hosted machine learning models for drug identification. A near-infrared handheld device (TactiScan) scans suspected samples, and the data is analysed by machine-learning drug detection models hosted in the cloud. Police intercept approximately 10,000 suspected drug samples per year.

Police have also described a set of tools aimed at online child exploitation investigations, including specialist image and video analytics tools, such as Griffeye, to identify child sexual abuse material, through image and video analytics including hash matching and facial comparison. Griffeye can identify and match images and videos depicting different suspects and victims directly on import, allowing investigators to identify unique individuals, narrow and prioritise material for review, and reduce investigator exposure to harmful content.

Finally, in 2025 New Zealand Police ran a simulation of a mock criminal trial at the Auckland High Court, to test TraceLex, a tool deployed to identify suspects by tracking their online activity. The mocktrial case explored the admissibility of evidence obtained through warrants relying on AI analysis.

Prosecutors  

New Zealand’s prosecution system is split between Police Prosecutors and Crown Solicitors or Crown Prosecutors.

1. Police Prosecution Service prosecutors conduct proceedings for minor criminal and traffic prosecutions commenced by New Zealand Police, except where the proceeding becomes a Crown prosecution.
2. Crown prosecutions, including all prosecutions in the High Court and all jury trials, are conducted by Crown Solicitors, who are private practitioners appointed for a particular district and supervised by the Solicitor-General. Because Crown prosecutions are outsourced to private firms, there is no single, centrally mandated AI technology publicly identified as being used across all Crown Solicitors/Crown Prosecutors. AI tool use is therefore likely to vary by firm and district.

As at June 2026, there is no reported use of AI by Police Prosecution Service prosecutors or Crown prosecutors. New Zealand’s regulatory framework (see below) imposes strict limitations on AI use by prosecutors.

Prosecutorial offices have, however, acknowledged the potential benefits of using AI, including for achieving ‘efficiency gains’ such as streamlining core processes, decision-making support, and evidence integrity. But they have also acknowledged privacy concerns and deepfake issues.

Courts

As at June 2026, there are no reported AI uses by New Zealand courts.

However, New Zealand courts are in the process of developing a digital case management system called Te Au Reka to transform the administration of courts in New Zealand. Te Au Reka is a digital caseflow management system which will assist in logistics management, content management, procedure management and administration management. As at June 2026, Te Au Reka is still under development and little is known about the extent to which it will deploy or rely on AI. Public Ministry of Justice Material published in May 2026 indicates that Phase 1 rollout for the Family Court is planned to begin in Christchurch and Ashburton District Courts and parts of the national services teams on 23 November 2026, with the court’s portal then planned to go live from March 2027.

Defence

AI adoption varies by organisation and practitioner. As such, the extent of AI use by defence counsel is largely unknown.

Legal research, analysis and drafting support

In February 2026, a New Zealand judge questioned the validity of an AI-drafted apology from a defendant on trial. The Judge questioned whether the apology could genuinely demonstrate remorse, a potential mitigating factor in sentencing. The use of AI did not make the apology inadmissible, but may in due course create doubt in a judge’s assessment as to whether remorse was actually expressed by the individual.

Victims

Victims do not have separate party standing in criminal proceedings in New Zealand. Victims’ rights in the criminal process are governed by the Victims’ Rights Act 2002, which provides rights to information, consultation on key decisions (such as bail and parole), and the opportunity to present a victim impact statement during sentencing.

As at June 2026, there are no reports of victims using AI in criminal proceedings in New Zealand.

TRAINING

Training on the responsible use of AI in the New Zealand justice system is not yet uniform across all justice actors. There is no single, system-wide mandatory AI training programme. Instead, training takes different forms across the judiciary, the legal profession, law enforcement, and the broader public service.

Law enforcement staff must complete mandatory training before being granted access to Microsoft Copilot Chat, the general-purpose generative AI tool now in active use across New Zealand law enforcement agencies. The Police’s Acceptable Use of Generative AI policy (see below) governs staff use of approved generative AI tools, and the Technology Capabilities List notes that staff who access specific platforms ‘are required to follow guidelines to use the service safely and effectively’. Internal training on specific investigative tools, such as BriefCam, Griffeye, and Cellebrite appears to be organisation- and role-specific rather than standardised across all staff.

At the broader institutional level, the Evidence Based Policing Centre completed two AI evidence reviews in early 2024 (available here and here), which function as awareness-raisias ofng resources for Police leadership and staff on the opportunities and risks of AI in law enforcement. As at June 2026, training does not appear to cover deepfakes specifically as a standalone module.

In August 2025, New Zealand Police organised a series of mock court hearings in the Auckland, Christchurch and Wellington High Courts to test the admissibility of AI-derived evidence in a criminal justice context. These exercises, involving a real judge and real lawyers, served as training for justice sector participants on the evidentiary and procedural challenges posed by AI tools, including “black box” algorithms, the absence of peer review, and the reliability of AI-generated analysis as the basis for search warrants.

Police Prosecutors are subject to the Police’s internal training requirements, including the mandatory training for Copilot Chat access (see above). As at June 2026, there is no publicly identified prosecutor-specific AI training programme.

Crown Prosecutors (as lawyers) and defence lawyers have access to various optional training programmes and workshops on AI, but there are no public reports of mandatory training. The New Zealand Law Association has offered various webinars on generative AI in legal practice, including a 2026 webinar on ‘Navigating the Law with Generative AI’. The New Zealand Law Society’s continuing legal education provider offers a range of professional development courses, including occasionally AI-specific courses.

For judicial actors, the Judges Association of New Zealand maintains an ‘AI and tech’ resource section and publishes articles covering topics including deepfakes and digital evidence, AI hallucinations in legal filings, and emerging uses of AI in courts. However, as ofJune 2026, there is no publicly available information indicating that a mandatory AI training programme has been established for judges or judicial support staff.

As at June 2026, there is no specific training available, across all actors in the criminal justice system, on deepfakes.

REGULATION

As at June 2026, there is no overarching regulation governing the use of AI in the justice system of New Zealand. Broader AI policy follows a ‘light-touch, principles-based approach’, relying on existing legal frameworks, including privacy, human rights, evidence and professional obligations, while encouraging practical adoption.

As such, New Zealand’s approach to AI in criminal proceedings and the justice system is currently governed by (1) technology-neutral legal frameworks (existing rules of procedure, evidence, privacy, human rights); and (2) sector-specific guidance and policies issued by the judiciary, Police and the Law Society.

While, in theory, a number of existing statutes specifically allow particular government departments and public entities to utilise ‘automated electronic systems’ for decision-making (for example, the Social Security Act 2018, Customs and Excise Act 2018, and the Summary Proceedings Act 1957), as at June 2026, there is no evidence that such systems are in use or are under consideration.

Guidelines for practitioners

As at January 2026, New Zealand has not formally adopted the UNESCO Guidelines for the Use of AI Systems in Courts and Tribunals (2025) as binding court rules or a mandatory protocol. However, the New Zealand judiciary has expressly engaged with UNESCO’s work: the Chief Justice’s 2024 Annual Report notes that, following the publication of New Zealand’s own generative AI guidelines, UNESCO published Draft Guidelines for consultation in 2024 and states that these ‘should be of interest to judges, legal practitioners, and all agencies working in the justice area.’

Nonetheless, a number of guidelines have been issued for practitioners across New Zealand, governing AI use by lawyers, courts, and law enforcement.

New Zealand Law Society, ‘Lawyers and Generative AI’ (2024)

AI use by New Zealand lawyers is governed at the professional level by Law Society guidance. The ‘Lawyers and Generative AI’ Guidance (2024) identifies key professional risks and controls, including accuracy and hallucinations, confidentiality and privilege, privacy and cybersecurity, intellectual property, supervision and competence. The guidance is accompanied by a practical checklist for lawyers and firms. In particular, the Law Society guidance provides:

The guidance is accompanied by a practical checklist covering purpose identification, due diligence on vendors, data input management, regulatory compliance, cybersecurity, and exit planning.

Courts of New Zealand, Guidelines for Use of Generative Artificial Intelligence in Courts and Tribunals (2023)

On 7 December 2023, the Courts of New Zealand issued Guidelines for Use of Generative Artificial Intelligence in Courts and Tribunals. The judiciary has published three generative AI guidelines for: (1) judges, judicial officers, tribunal members and judicial support staff; (2) lawyers; and (3) non-lawyers, including self-represented litigants, McKenzie friends and lay advocates.

Police Guidelines

AI use by NZ Police is governed by the New Technology Framework and the Police’s Acceptable Use of Generative AI (genAI) Policy. Its headline requirements are:

• That staff complete genAI training before use;
• That only approved tools may be used, and only within their approved scope;
• Fully automated use is not permitted: a human must remain in the loop, and the user stays accountable for outputs;
• Users must critically evaluate outputs for bias, inaccuracy, and hallucination, and manage privacy and information-security risks;
• Outputs should be labelled as AI-generated and identify the tool used, with the label carried through on any further distribution.

Moreover, the Policy establishes that permitted uses are set by a risk-based assessment under a Technology Assurance Framework, with limited authority for urgent one-off use at senior level. The use of genAI outputs in any court context is not permitted unless the court has been engaged and has approved it, and transcription or translation outputs may not be used directly as evidence without human review. Vendors may use genAI only with Police’s agreement and remain accountable for their outputs.

Criminal procedure rules

As at June 2026, no AI-specific amendments have been introduced to New Zealand’s core criminal procedure and evidence statutes. Criminal proceedings remain governed by the Criminal Procedure Act 2011 and the Criminal Procedure Rules 2012, which set the procedural framework for charging documents, case management, and pre-trial applications (including applications under the Evidence Act, see below). Questions about AI-generated or AI-analysed material therefore fall to be determined under New Zealand’s Evidence Act 2006’s existing admissibility architecture. These statutes do not expressly require the disclosure of AI tools.

The Evidence Act 2006 was last reviewed by Te Aka Matua o te Ture Law Commission in 2024 (NZLC R148). The review focused on operational issues in evidence law but, perhaps surprisingly, did not include any AI-specific analysis or recommendations. A provision in the Evidence Act requiring 5-yearly reviews of the Act was repealed in 2022 and there are currently no plans for further reviews to address AI-related issues.

Under the Evidence Act, relevant evidence is admissible (section 7) unless its probative value is outweighed by the risk of unfair prejudice or needless delay (the general exclusion in section 8), subject to specific rules for particular categories. For AI-generated or AI-analysed material, the provisions most likely to be engaged are the expert-opinion test in section 25 (expert evidence is admissible only where it offers the fact-finder substantial help), the hearsay rules, and section 30 (discretionary exclusion of improperly obtained evidence).

More broadly, existing criminal offences are capable of applying to the use of AI to fabricate or tamper with evidence in court proceedings. Under the Crimes Act 1961, it is an offence to fabricate evidence (section 113, carrying a maximum penalty of seven years’ imprisonment) or to attempt to obstruct, prevent, pervert, or defeat the course of justice (section 116). These general offences apply regardless of whether fabrication is carried out manually or through the use of AI tools. As addressed below, the Supreme Court has recently warned that reliance on false AI-generated citations may amount to obstruction of justice or contempt of court.

Data protection legislation

New Zealand’s primary data protection statute is the Privacy Act 2020, which applies broadly to agencies (including public-sector and private bodies) handling personal information. The Privacy Act applies to personal information held by law enforcement, prosecutors, defence counsel and the courts. It sets out 13 Information Privacy Principles (‘IPPs’) governing collection, storage, use and disclosure, including:

New Zealand also has a Biometric Processing Privacy Code 2025, made under the Privacy Act 2020. The Code came into force on 3 November 2025 for new biometric processing, with a grace period until 2 August 2026 for existing biometric processing. It is relevant to AI-enabled justice tools that process biometric information, including facial recognition and biometric matching, because it imposes more specific privacy obligations for biometric information than the general IPPs.

Cybersecurity laws

The Harmful Digital Communications Act 2015 (HDCA) addresses online harassment and harmful communications by providing a civil pathway to pursue perpetrators of online harm and criminal offences for the most serious conduct (e.g., deliberately causing serious emotional distress through harmful posting). This is likely to come into play when AI is used to generate or amplify harmful content (e.g., deepfakes).

Human rights

Human-rights constraints relevant to AI use and algorithmic and AI-assisted decision-making in the criminal justice system arise primarily from:

1. Statutory rights under the New Zealand Bill of Rights Act 1990 (which dovetails with and affirms New Zealand’s commitments under the International Covenant on Civil and Political Rights (‘ICCPR’));
2. Domestic anti-discrimination law under the Human Rights Act 1993;
3. Statutory privacy protections under the Privacy Act 2020 (including in relation to biometric information); and
4. Compliance with Te Tiriti o Waitangi or the Treaty of Waitangi (which is regarded as a major source of New Zealand’s constitution and a founding document whose principles are enshrined in a number of statutes and which policy-makers are required to consider).

Fair trial and privacy guarantees under other international human rights treaties to which New Zealand is a party may also be relevant. In particular, articles 37 and 40 of the Convention on the Rights of the Child – which protect children’s rights in the administration of juvenile justice, including the right to be treated with dignity, the right to legal assistance, and the right to have their privacy respected – will be engaged where algorithmic tools are used in decisions affecting children and young people.

Algorithmic tools that influence policing priorities, risk assessment, surveillance intensity, or operational decision-making can therefore create (or amplify) inequities if they embed biased data, imperfect proxies, or unexamined assumptions.

The Algorithm Charter for Aotearoa New Zealand directly links responsible algorithm use to Treaty commitments. It requires agencies to manage algorithms in ways that ‘prevent unintended bias’ and reflect Treaty principles, and includes commitments to ensure data is fit for purpose and to identify and manage bias.

New Zealand Bill of Rights Act 1990 and Human Rights Act 1993

In the criminal context, the most directly relevant protections in the New Zealand Bill of Rights Act 1990 include:

1. Freedom from discrimination (Section 19), which incorporates the prohibited grounds in the Human Rights Act 1993;
2. Protection against unreasonable search or seizure (Section 21), which is engaged by many forms of technology-enabled surveillance and data extraction; and
3. Minimum standards of criminal procedure (Section 25), which protects core fair-trial guarantees (including the right to a fair and public hearing, to be presumed innocent, and to examine witnesses), and is potentially engaged where, in future, AI systems may come to shape evidential disclosure, investigative leads, or the reliability of material presented in proceedings.

Alongside the Bill of Rights, the Human Rights Act 1993 provides the principal statutory framework prohibiting discrimination across specified grounds (e.g. race, gender, sexuality, age, marital status, etc.) and supports enforcement mechanisms relevant to public decision-making that produces discriminatory impacts.

Te Tiriti o Waitangi - The Treaty of Waitangi 1840

Te Tiriti o Waitangi or the Treaty of Waitangi is regarded as a major source of New Zealand’s constitution and, in many respects, law. While the text of Te Tiriti is not directly enforceable as ‘law’ in all contexts, its principles have been developed through legislation, the courts and the Waitangi Tribunal, and are applied in many policy and operational settings affecting Māori rights and interests. Māori are disproportionately represented in all parts of the criminal justice system.

Soft law instruments

Finally, ‘soft law’ instruments used across parts of the New Zealand public sector reinforce these human-rights themes. For example, New Zealand’s Algorithm Charter commits participating agencies to ensuring algorithms are consistent with law and human rights values and to considering impacts on Māori and other groups (including by reference to Te Tiriti principles).

Operational policies for specific technologies can also build human-rights considerations explicitly. For example, New Zealand Police’s Policy on the use of facial recognition technology emphasises the need for human oversight for use to have regard to privacy and human-rights risks (including bias and potential disproportionate impacts).

Outlook

At the central government level, New Zealand’s near-term direction is to encourage AI uptake while relying on existing, technology-neutral legal frameworks and sector guidance to manage risks, rather than creating a comprehensive, AI-specific statute for the justice sector. This approach is reflected in the Government’s first national AI strategy, New Zealand’s Strategy for Artificial Intelligence: Investing with Confidence (Ministry of Business, Innovation and Employment, July 2025), which recommends voluntary guidance intended to build confidence and reduce regulatory uncertainty.

In the justice system, the same balancing exercise seen in other jurisdictions is likely to guide the uptake of AI: harnessing AI’s potential benefits (such as improved access to justice and administrative efficiency) while managing risks including bias, privacy and surveillance impacts, and over-reliance on tool outputs (automation bias).

The most active deployment of AI technologies is likely to remain concentrated in policing. New Zealand Police’s public materials indicate active work to strengthen governance for ‘emergent technologies’, including publication of a Technology Capabilities List and the development of internal policy and research programmes (including the Evidence Based Policing Centre’s two 2024 horizon-scanning reviews of AI opportunities and risks across Five Eyes countries).

A significant limiting factor for generative AI use for legal work in New Zealand is the relatively small size and distinctiveness of the New Zealand legal and cultural context. The judiciary’s Guidelines for Use of Generative Artificial Intelligence in Courts and Tribunals (2023) caution that currently available generative AI tools have had limited access to New Zealand-specific legal training data, and ‘generally do not account for New Zealand’s cultural context’, including Māori and Pasifika values and practices. As such, generative AI tools may generate culturally unattuned content. These constraints are likely to shape both uptake of AI in the New Zealand justice system and the kinds of uses regarded as appropriate. Although demand for New Zealand-specific datasets may support the development of local tools over time, the relatively small scale of the legal market (around 17,000 practising lawyers) is likely to limit the business case for building and maintaining bespoke New Zealand-trained systems.

Cases

Data protection

In Tamiefuna v. R [2025] NZSC 40 the Supreme Court considered whether photographs taken by Police during a random traffic stop and then retained or used for later identification purposes could be used to convict a person of an unrelated crime. The case began with a traffic stop by a police officer in 2019, during which an officer took pictures of Mr Tamiefuna with his phone and uploaded the images to the national intelligence database. The photo of Tamiefuna on the police database matched CCTV footage taken three days earlier after an aggravated robbery.

The Court allowed the appeal, quashed the conviction, and ordered a retrial, holding (in substance) that the photographs were obtained through an unlawful and unreasonable ‘search’ (New Zealand Bill of Rights Act 1990, s 21), the resulting evidence should be excluded under the Evidence Act 2006 (s 30). The Court found the taking of the photo was unlawful and unreasonable because the officer was not investigating any specific crime when he took it. Uploading the photo to the database and keeping it there was also unlawful and unreasonable.

Although not an AI case as such, the decision concerns the collection and retention of image data in police intelligence systems and the downstream use of that data for later investigative matching. A particularly important theme in the judgment is the Court’s concern about police officers collecting identification material without a specific investigative purpose (i.e., photographing someone ‘just in case’ it becomes useful later). Though the case did not involve the use of facial recognition software, it scrutinises the legality of creating and retaining image-based identification data in police intelligence systems – an enabling step for later biometric matching.

Misuse of AI in court filings

In Jones v. Family Court at Whangārei [2026] NZSC 1 (Supreme Court, 11 February 2026), the Supreme Court addressed the misuse of AI-generated material in court submissions. The case itself concerned a self-represented applicant seeking leave to appeal in a Family Court care-of-children dispute. In declining leave, the Court noted that the applicant’s submissions ‘cited a number of authorities which appear to have been hallucinated by an Artificial Intelligence (AI) application.’

The Supreme Court stated that: ‘[m]isuse of AI in legal proceedings has serious implications for the administration of justice and public confidence in the justice system’ and that ‘[p]ersons filing submissions in court must ensure all authorities referred to are genuine and correctly cited.’ The Court pointed to the Courts of New Zealand’s guidelines on AI use for non-lawyers and went further by warning that ‘[r]eliance on false citations, including the unverified outputs of AI applications, may in serious cases amount to obstruction of justice or contempt of court’ – referencing the Crimes Act 1961 and the Contempt of Court Act 2019.

Jones is the first New Zealand Supreme Court decision to directly address AI-generated hallucinations in court submissions. While the warning arose in the context of a self-represented litigant, the Court framed the issue in terms of the integrity of the administration of justice – principles that apply equally to lawyers.

The Employment Relations Authority has also encountered AI-generated content in tribunal proceedings. In QTR v. BXD [2025] NZERA 716, the Authority found that an employee had used a generative AI platform to assist with preparing responses during an Authority investigation. The Authority identified multiple issues, including the inclusion of hallucinated legal cases, incorrectly cited authorities, and the uploading of confidential and personal workplace information into an AI platform – raising both accuracy and privacy concerns. The Authority issued a reminder that information provided by generative AI must be checked before being relied on in tribunal and court proceedings, referencing the Courts of New Zealand’s guidelines on generative AI use.

In Yorston v. Attorney-General [2026] NZCA 15, the Court of Appeal considered whether errors in computer-generated criminal records produced by the Ministry of Justice’s case management system were amenable to judicial review. A software issue had caused the appellant’s Convictions History Report to conflate conviction dates with sentencing dates, and his criminal and traffic history report contained incorrect offence dates.

The High Court had declined judicial review, holding that the production of a Convictions History Report was ‘wholly administrative’ and ‘essentially mechanical’ and therefore not the exercise of a statutory power of decision amenable to judicial review. While the Court of Appeal dismissed the appeal as moot, the specific errors having been corrected, it expressly rejected the proposition that the automated nature of a system places it beyond judicial scrutiny. The Court held that Convictions History Reports are generated by systems adopted, managed and operated under the administration of the Ministry of Justice, and that the Ministry bears ultimate responsibility for the accuracy of those systems. The production of Convictions History Reports was found to result from the exercise of executive powers of government and must therefore be amenable to review. The Court also flagged broader concerns about the reliability of the case management system, noting that criminal and traffic history reports are relied on by sentencing courts daily and that it was ‘concerning’ that both report types contained inaccuracies.

Although not an AI case as such, Yorston is significant for the AI and algorithmic justice context because it establishes that government agencies cannot shield automated decision-making systems from judicial oversight by characterising their outputs as merely ‘mechanical’ or administrative. The decision affirms that where an agency exercises control over an automated system that produces records with legal consequences, including for sentencing, employment vetting and other justice processes, the agency retains responsibility for the accuracy of those outputs and the exercise of that power remains subject to judicial review.

Right to privacy

A series of District Court rulings in late 2024 addressed challenges to police reliance on Automated Number Plate Recognition hits obtained through Auror (see above), with defendants arguing that retrospective access to Automated Number Plate Recognition-derived location or movement data should require a warrant or production order and/or should be treated as an unlawful search. The District Court rulings rejected these challenges, treating ANPR-enabled access to camera data as akin to leveraging ubiquitous CCTV in public spaces and stating (in at least one ruling) that there was ‘no privacy interest’ in the ANPR data in that context.

However, appeals are underway. In September 2025, a Court of Appeal pre-trial hearing began in Wellington arising from an appeal to challenge Police reliance on Auror-derived Automated Number Plate Recognition evidence, with appellants arguing the tool enables mass surveillance and bypasses traditional controls (warrants/production orders). As at June 2026, the Court’s decision is pending.