AI Forum NZ — Generative AI Working Group A living document
Living Whitepaper
Archived

You're reading an archived edition. Read the latest version from 1 September 2026.

AI Adoption in Healthcare in New Zealand: A Living Whitepaper

Updated: 13 July 2026

Executive Summary

AI adoption in New Zealand healthcare is still best characterised as selective operational deployment under tight clinical and governance controls, rather than broad automation. The most mature scaled public-sector use remains ambient documentation: as of 28 February 2026, AI scribe technology was live in every emergency department nationwide for 1,250 frontline staff, with more than 1,000 additional licences being progressed, largely for mental health teams. Health New Zealand’s privacy statement continues to say AI is used to assist clinicians, that many uses remain limited or pilot-phase, and that AI is not used to make automated decisions about a person’s healthcare. (beehive.govt.nz)

The biggest developments since the previous update on 10 June 2026 are less about new nationwide rollouts and more about implementation maturity. On 11 June 2026, the Health Research Council announced a $5 million New Zealand-led ICU trial that will test machine-learning-guided oxygen therapy across 50 ICUs in New Zealand and Australia and more than 24,000 patients. On 26 June 2026, a New Zealand Medical Journal paper on AI-enabled diabetic retinal screening published a grounded local implementation lesson: promising AI use cases still run into workflow, system integration, clinician-readiness, and tool-fit barriers. On 8 July 2026, Health NZ announced a partnership with UCLPartners in the United Kingdom to strengthen digital and AI innovation, including work relevant to mental health, diagnostics, and data integration. (hrc.govt.nz)

Breast screening remains the clearest next-wave clinical deployment. The latest official milestone is still the 27 May 2026 announcement that procurement is underway for a preferred AI mammogram-reading tool, with testing and validation ahead of a planned rollout from early 2027. Publicly, there has not yet been a later announced implementation milestone beyond that procurement step. (beehive.govt.nz)

The broader picture is now clearer: New Zealand is still early in health-system-wide AI adoption, but it is no longer simply experimenting. The public record points to a model of clinician-supervised augmentation, concentrated first in documentation and workflow support, with imaging, screening, and decision support moving forward more cautiously through validation, governance, and research. That is an inference drawn from Health NZ guidance, Ministry positioning, the current deployment pattern, and the June–July 2026 research and policy updates. (info.health.nz)

What Has Changed Since 10 June 2026

  • Clinical AI research moved closer to frontline decision support. On 11 June 2026, the HRC announced the REVOLUTION trial, a New Zealand-led programme grant of nearly $5 million to test AI-guided oxygen therapy in ICU care across 50 ICUs in New Zealand and Australia, enrolling more than 24,000 patients. (hrc.govt.nz)
  • New local implementation evidence arrived for screening AI. On 26 June 2026, NZMJ published a paper on the diabetic retinal screening use case, concluding that implementation barriers included digital systems, workflow redesign, clinician readiness, and suitability of the AI tools themselves. (pubmed.ncbi.nlm.nih.gov)
  • Health NZ added an international AI innovation partnership. On 8 July 2026, Health NZ said its HealthX programme had entered an agreement with UCLPartners, with cited relevance to AI scribes, mental health settings, AI-enabled diagnostics, and better integration of patient information across systems. (healthnz.govt.nz)
  • The national shared-record programme slipped but became more explicit about security and timing. In Health NZ’s July 2026 update, the Shared Digital Health Record rollout timeline was extended for additional information-sharing checks; immunisation and medication data are now expected later in 2026, while access to primary care data is expected from mid-2027. (healthnz.govt.nz)
  • No new public milestone has overtaken the breast-screening procurement announcement. As of 13 July 2026, the latest official public step remains procurement and validation planning for AI mammogram reading ahead of a proposed early 2027 rollout. (beehive.govt.nz)

Current State of AI Adoption

1. Public hospitals: documentation support remains the most mature live deployment

The clearest scaled deployment in the public hospital system remains the emergency department AI scribe rollout announced on 28 February 2026. The Government said the tool was live in all emergency departments, reaching 1,250 ED doctors and frontline staff, with over 1,000 more licences being progressed, predominantly for mental health teams. (beehive.govt.nz)

This is still strategically important because it shows where Health NZ has been most willing to scale AI: embedded documentation support with human review, not autonomous clinical decision-making. Health NZ’s privacy statement says AI-generated information that could affect the clinical record or decision-making is reviewed by the responsible clinician, and that AI is not used to make automated decisions about care. (info.health.nz)

Health NZ is also signalling interest in AI at the system “front door,” but still at an exploratory stage. Its Symptom Checker programme says Health NZ is exploring an AI-supported online symptom checker to help people navigate to the right health service, while explicitly stating that it would support, not replace, clinical decision-making and that a clinically led governance group is being set up. (healthnz.govt.nz)

2. Primary care: adoption is fast, uneven, and increasingly integrated into local software

Primary care remains the fastest-moving non-hospital setting. A New Zealand survey paper published on 26 March 2026 found that among 197 respondents, 40% had experience with AI scribes; reported benefits included reduced multitasking, time savings, and improved rapport, but the study also found gaps in consent, terms-and-conditions awareness, and confidence in legal and ethical compliance. (pubmed.ncbi.nlm.nih.gov)

A separate workforce signal from the Royal New Zealand College of General Practitioners is consistent with that direction of travel: its current workforce survey page says 41% of GPs and 16% of rural hospital doctors had already used, or intended to use, AI, with note-taking and scribe tools the most common use case. (rnzcgp.org.nz)

Vendor offerings are also becoming more integrated. Medtech says its Medtech AI platform, launched in New Zealand on 23 February 2026, is built into Medtech Evolution, synthesises patient history from the PMS, stores data locally in New Zealand, does not retain consultation audio, and requires clinicians to confirm explicit patient consent before recording starts. (medtechglobal.com)

The implication is that primary care adoption is moving from isolated experimentation toward workflow integration, governance, and clinician onboarding, but not yet to a standardised national operating model. That is an inference supported by the survey data, the RNZCGP workforce signal, and the PMS-integrated product direction. (pubmed.ncbi.nlm.nih.gov)

3. Aged care: AI scribes are spreading beyond hospitals and GP clinics

Aged residential care is now a clearer part of the adoption story. On 12 May 2026, Metlifecare said it would roll out the HEIDI ambient scribe tool across three more care homes after a four-month trial at Parkside Village involving 12 registered nurses. Nurses reported less time spent on documentation, better continuity of care, and improved handovers, while Metlifecare said clinicians remained responsible for reviewing, editing, and approving all documentation. (metlifecare.co.nz)

This matters because it shows that AI documentation tools are not confined to acute hospital environments. They are now being deployed in settings where staffing pressure, handover quality, and documentation load are also operational pain points. (metlifecare.co.nz)

4. Screening, imaging, and diagnostics: still the strongest next-wave clinical lane

Breast screening remains the most visible national clinical AI programme in the pipeline. The Government said on 27 May 2026 that procurement was underway for a preferred AI mammogram-reading tool for testing and validation, ahead of a planned rollout from early 2027. The proposed model would have AI perform one of the two independent reads in the existing screening process, while clinicians remain central to diagnosis and follow-up decisions. (beehive.govt.nz)

The operational case is significant: the same announcement said around 270,000 women aged 45 to 69 are screened annually through BreastScreen Aotearoa, with a phased age extension to 74 underway. That creates both rising volume and workforce pressure. (beehive.govt.nz)

Retinal screening remains the other important imaging-related use case. The new NZMJ paper on diabetic retinal screening argues that the New Zealand proof-of-concept showed AI implementation is not a simple plug-in exercise; success depends on system readiness, workflow redesign, resourcing, and organisational support. Separately, Health NZ’s National Diabetes Roadmap 2026 includes a specific action to evaluate the role of AI in supporting retinal photoscreening. (nzmj.org.nz)

Health NZ’s privacy statement reinforces that imaging AI is already part of the live but limited-use landscape, saying AI is used to read, analyse, or review clinical information for preliminary results, including images, x-rays, scans, or mammograms, while noting many such activities are still being tested on a limited basis. (info.health.nz)

Governance, Regulation, and Digital Foundations

1. The official operating model is still “assist clinicians, do not automate care”

Health NZ’s privacy statement says AI operates in a closed environment, that personal information is not used to build commercial AI or third-party generative models, and that AI is used for tasks including transcription, consultation summaries, document generation, staff knowledge tools, and preliminary review of images and scans. It also states that many of these activities are still in pilot phase and that Health NZ does not use AI to make automated decisions about a person’s healthcare. (info.health.nz)

Health NZ’s guidance on generative AI and large language models, last updated in May 2026, is equally explicit: staff must not enter patient or other sensitive information into unapproved LLMs, must not use LLMs for clinical decisions or personalised advice, and Health NZ says it does not currently have any private LLMs available. (healthnz.govt.nz)

The Medical Council’s AI guidance, published on 10 March 2026, sets the professional standard alongside that system stance. It says doctors remain responsible for all clinical decisions and actions, must be satisfied the AI tool is safe and suitable, must obtain informed consent in some situations including some recording use cases, must consider bias and equity, and must not use AI in ways that impersonate a doctor. (mcnz.org.nz)

2. Governance is becoming more operational and standardised

Health NZ’s National Artificial Intelligence and Algorithm Expert Advisory Group (NAIAEAG) remains the key governance mechanism. Health NZ says all AI development or implementation plans must be registered with the group, which oversees ethical, technical, clinical, and operational standards. Its current terms also say brief public summaries of advice or approvals will be made available and that a National Register of AI in use within Health NZ will be made publicly available. (healthnz.govt.nz)

That governance stack is being reinforced by a New Zealand-specific evaluation framework. The University of Auckland’s TRANSFORM programme says that in March 2026 the framework was endorsed by Health NZ’s Executive Leadership Team and is now being used to assess all proposed AI tools across Health NZ services. (transform.auckland.ac.nz)

3. Privacy and cyber security are now central to the adoption story

The privacy baseline tightened on 1 May 2026. The Office of the Privacy Commissioner says the Health Information Privacy Code 2020 was amended in March 2026 to reflect new IPP3A requirements, with the updated code in force from 1 May 2026. (privacy.org.nz)

Cybersecurity has also become inseparable from digital and AI scale. On 27 May 2026, the Ministry of Health said an independent review into the Manage My Health cyber incident involved the theft of highly sensitive health information affecting 99,000 people and concluded stronger system stewardship and third-party assurance were needed. On 28 May 2026, the Government announced $153.6 million for Health NZ to expand national cyber monitoring, strengthen data security processes, and deliver critical IT safety upgrades. (health.govt.nz)

For AI adoption, the significance is straightforward: the more healthcare depends on connected data, external vendors, and digitally mediated workflows, the more cyber resilience becomes a prerequisite for trust and scale. That is an inference, but it is strongly supported by the MMH review findings, Health NZ’s privacy posture, and the Budget 2026 cybersecurity investment. (health.govt.nz)

4. The digital backbone is still being built

Health NZ’s July 2026 Shared Digital Health Record update said the rollout timeline had been extended to allow additional checks under the Government Digital Delivery Agency information-sharing standard. Immunisation and medication data are now expected later in 2026, while access to primary care data is expected from mid-2027. (healthnz.govt.nz)

That matters for AI because many higher-value use cases depend on clean, secure, shareable longitudinal data. Health NZ’s July update also tied the programme more directly to sector-wide information-sharing and security awareness work. (healthnz.govt.nz)

Health NZ’s new agreement with UCLPartners, announced on 8 July 2026, adds an international benchmarking layer to this foundation-building work. Health NZ said the partnership will support evidence-led assessment and scaling of digital and AI innovation, with current relevance to emergency-department AI scribes, mental health AI, diagnostics, and bringing patient information together across systems. (healthnz.govt.nz)

Research and Innovation Pipeline

New Zealand’s research pipeline remains broader than its live deployment footprint. The HRC said in August 2025 it had invested $4.6 million across 10 AI in Healthcare studies, and that portfolio is still shaping the implementation agenda. (hrc.govt.nz)

The portfolio is notably practical:

  • Critical care: the REVOLUTION trial will test machine-learning-guided oxygen therapy in ICU care across New Zealand and Australia. (hrc.govt.nz)
  • Radiology: an HRC-funded project is evaluating whether AI can improve the speed and accuracy of chest X-ray interpretation and reporting in New Zealand hospitals, while also studying workflow effects and implementation barriers. (hrc.govt.nz)
  • Retinal screening: another HRC-funded project focuses on evaluating performance and implementation of AI-enabled diabetic retinopathy screening in Health NZ Waitematā. (hrc.govt.nz)
  • Digital pathology: Otago-led work is building AI-enhanced pathology tools for gastrointestinal cancers to improve treatment selection and avoid unnecessary surgery. (hrc.govt.nz)
  • Heart failure: Auckland researchers are studying how an AI-derived management support tool could work within secure digital health systems for chronic heart failure care. (hrc.govt.nz)
  • Postoperative monitoring: Auckland-led work is using digital tools and AI to detect deterioration earlier after surgery, with explicit attention to equity, privacy, and health data sovereignty. (hrc.govt.nz)
  • Mental health ethics and data justice: separate HRC-funded projects are focused on safe, equitable AI use in youth mental healthcare and on data justice frameworks for AI in health in Aotearoa. (hrc.govt.nz)

The pattern here is consistent: New Zealand’s strongest AI pipeline is concentrated in clinically bounded, measurable, supervised use cases rather than open-ended automation. (hrc.govt.nz)

Trust, Equity, and Social Licence

Trust remains a binding constraint on adoption. A February 2026 NZMJ viewpoint argued that patient trust is central to AI implementation in Aotearoa New Zealand and highlighted issues including public benefit, governance, data protection, meaningful choice, clinician responsibility, and Māori representation. (nzmj.org.nz)

That concern is echoed by community-specific research. A 2026 study on Pasifika perspectives on AI and asthma management found interest in AI’s potential, but also recurring concerns around privacy, accuracy, access, age-related digital divides, and design suitability. (openrepository.aut.ac.nz)

Health NZ’s own LLM guidance aligns with those concerns by explicitly warning about privacy breaches, inaccurate information, inequities and bias, lack of transparency, Māori data sovereignty, and weak support for te reo Māori and other minority languages. (healthnz.govt.nz)

1. Documentation-first adoption is now established across multiple care settings

Emergency departments, general practice, and aged residential care all show live or active adoption of ambient documentation tools. (beehive.govt.nz)

2. Imaging and interpretation remain the strongest near-term clinical scale-up lane

Breast screening procurement, retinal-screening evaluation, chest X-ray implementation research, and pathology projects all point in the same direction. (beehive.govt.nz)

3. Research is moving from exploratory study toward implementation science and trials

The June ICU trial announcement and the June retinal-screening paper both show a system that is asking not only “does the model work?” but “can this be governed, integrated, and scaled safely?” (hrc.govt.nz)

4. Governance has become an operating system, not just a principles discussion

NAIAEAG registration, the national evaluation framework, Medical Council guidance, Health NZ LLM restrictions, and the updated privacy code together form a much denser governance stack than New Zealand had a year ago. (healthnz.govt.nz)

5. Cybersecurity and data-sharing capability are now pacing functions for AI adoption

The MMH breach review, Budget 2026 cyber investment, and delayed Shared Digital Health Record rollout all show that secure interoperability is as important as model performance. (health.govt.nz)

6. International benchmarking is increasing, but local validation remains the rule

Health NZ’s UCLPartners agreement points to more structured learning from overseas systems, while breast-screening procurement and the retinal-screening case both reinforce New Zealand’s insistence on local testing, validation, and contextual fit. (healthnz.govt.nz)

Overall Assessment

As of 13 July 2026, AI adoption in New Zealand healthcare has advanced to a stage best described as targeted operational use with stronger implementation discipline. The live deployments with the clearest scale are still in documentation and workflow support; the most visible national clinical programme remains AI mammogram reading for BreastScreen Aotearoa; and the most important new developments since 10 June 2026 are the move toward larger clinical trials, sharper implementation learning, and stronger digital-and-governance infrastructure. (beehive.govt.nz)

The public evidence does not yet support a view that New Zealand has moved into broad autonomous clinical AI. Instead, it supports a narrower conclusion: the country is building an AI model based on supervised augmentation, phased validation, and tighter controls around privacy, equity, and cyber risk. That approach may look slower than more aggressive international markets, but it is increasingly coherent. (info.health.nz)

← Back to all articles ⚑ Flag an issue