AI in Academic Research

Executive Summary

• The evidence base has improved, but adoption is still not nationally measurable. The New Zealand Research Information System (NZRIS) went live on 23 September 2026, bringing together funding and research-activity information from major public funders. It is a significant measurement improvement, but it does not record how researchers use AI inside their workflows and is not yet comprehensive. mbie.govt.nz

• International AI research activity has moved into funded projects. New Zealand organisations are participating in Horizon Europe projects on marine AI-powered bioprospecting and distributed generative-AI agents. These are substantive research collaborations, but they are not evidence of broad operational AI deployment across universities. mbie.govt.nz

• University-level governance is becoming more operational. The University of Auckland’s requirement for doctoral candidates to discuss, document and declare GenAI use has been in force since 1 September 2026. It provides one of the clearest examples of AI governance being embedded in the research lifecycle. auckland.ac.nz

• New research hubs are treating AI as an interdisciplinary and societal issue. The University of Auckland launched Huanui: The AI Initiative, centred on human judgement, Māori and Indigenous perspectives, and collaboration between researchers, business, government and communities. auckland.ac.nz

• Research-to-implementation pathways are visible but remain narrow. University of Auckland spin-out RosterLab has entered a Health New Zealand pilot based on doctoral research into healthcare rostering. The pilot is operating in selected services, but its impact has not yet been independently evaluated. auckland.ac.nz

• The national AI platform remains unresolved in public documentation. NZIAT’s current AI page describes an AI Growth Engine being established, but does not identify a successful phase-two platform proposal. Government planning documents continue to describe the AI platform as an investment under development rather than a fully operating national centre. nziat.co.nz

• The strongest adoption signals remain domain-specific. Health imaging, robotics, bioeconomy research, climate science, software engineering and AI governance show concrete activity. Evidence of routine generative-AI use across academic disciplines, or of sector-wide productivity gains, remains thin.

What Has Changed Since the Last Update

NZRIS has created a new evidence layer

The most important system-level change is the launch of NZRIS on 23 September 2026. The platform brings together information from MBIE, Royal Society Te Apārangi and the Health Research Council, and its public interface currently lists approximately 15 funds, 5,500 grants, 665 organisations and 4,000 outputs. mbie.govt.nz

For AI in academic research, NZRIS could make it easier to identify publicly funded projects, collaborating organisations and emerging concentrations of capability. It may eventually support more systematic analysis of AI-related research funding across health, environment, agriculture, engineering and the social sciences.

Its limitations are important:

• The initial dataset is not a complete census of New Zealand research.
• Coverage depends on participating funders and research organisations.
• Data availability and classification vary between providers.
• NZRIS records reported administrative and descriptive information, not scientific quality, research impact or actual tool usage.
• The system does not contain research papers, raw datasets or detailed workflow records. nzris.govt.nz

NZRIS therefore improves visibility of the research system without yet solving the central AI-adoption question: how often researchers use AI, for which tasks, with what results and under what conditions.

Horizon Europe participation has produced new AI projects

Two newly listed Horizon Europe projects provide stronger evidence of active international research participation than the earlier pipeline announcements.

The four-year AIMARIA project, led by a European consortium with Victoria University of Wellington as the New Zealand partner, began on 1 September 2026. It will use AI and marine-microbiome data to support bioprospecting for medicines, enzymes and industrial compounds. MBIE has allocated NZ$606,913.21 in top-up funding to the New Zealand organisation. mbie.govt.nz

The CoAgent project, listed with a duration beginning on 1 October 2026, includes Auckland University of Technology and has received NZ$453,190.22 in MBIE top-up funding. It focuses on distributed, energy-efficient and trustworthy multi-agent generative AI operating across cloud, edge and physical environments. AUT researchers are contributing work on neuromorphic and brain-inspired agentic AI. mbie.govt.nz

These projects represent a step forward from proposal activity to contracted international research. They should still be distinguished from operational deployment:

• They are funded research programmes.
• Their expected outputs include methods, models, prototypes and research results.
• They do not show that these systems are running at scale in New Zealand universities or industry.
• The CoAgent project had a future start date of 1 October 2026 at the time of the latest public listing.

Huanui adds a human-centred research and engagement hub

The University of Auckland launched Huanui: The AI Initiative on 23 September. Based in the Business School, the initiative is intended to support research, teaching, innovation and public engagement rather than promote a particular commercial AI product. Its planned activities include an AI think tank, executive-in-residence roles, faculty champions and student AI partners. auckland.ac.nz

Huanui is notable because it frames AI capability as more than technical model development. Its stated approach places human judgement, purpose, whanaungatanga, kaitiakitanga, mahi tahi and rangatiratanga at the centre of research and engagement.

The initiative is still a newly launched institutional programme. There is not yet public evidence of its research outputs, sustained external partnerships or measurable impact. Its significance at this stage is strategic: it provides a visible institutional home for interdisciplinary AI research and translation.

A university spin-out has reached a live health-system pilot

RosterLab provides one of the clearest examples of academic AI research moving into an operating environment.

The company emerged from University of Auckland doctoral research into mathematical optimisation for complex hospital rosters. In July 2026, it confirmed a contract with HealthX, Health New Zealand’s digital innovation programme, for an initial pilot involving medical radiation technologist teams at Counties Manukau, Waikato, Starship and Waitematā. auckland.ac.nz

RosterLab reports that earlier work with Middlemore Hospital reduced roster-management time from approximately 120 hours per month to fewer than 20 hours. That result is company-reported and relates to prior work, not the current HealthX pilot. The University of Auckland also states that the pilot’s impact across participating services remains to be evaluated. auckland.ac.nz

The case is strategically important because it shows the full pathway from doctoral research to commercialisation and health-system testing. It also demonstrates why a funded pilot should not be described as a proven, scaled deployment.

The national AI platform still has no publicly named operator

The unresolved AI Research Platform remains the main point of continuity from the previous update.

NZIAT’s current AI page says that responsibility for the investment process moved to NZIAT in 2026 and that the organisation is establishing an AI Growth Engine. It does not identify a successful phase-two proposal or describe a commissioned national platform with user access, infrastructure, staff or operating metrics. nziat.co.nz

The Government’s Science Investment Plan 2026–2036 continues to refer to the AI platform as an investment under development and describes coordination between NZIAT and Research Funding New Zealand. mbie.govt.nz

The public record therefore supports a cautious conclusion: the national AI investment is committed in principle, but the operating centre of gravity has not yet been publicly established.

Current State of AI Adoption

Maturity assessment

AI is being used most credibly where the research problem is well defined

The strongest evidence concerns AI embedded in established research workflows rather than unrestricted use of general-purpose chatbots.

Current examples include:

• MRI analysis and prediction of stroke recovery.
• Computer vision and robotics for Māori sign-language interpretation.
• AI and brain-signal analysis for health assessment.
• Machine learning for climate and environmental modelling.
• Optimisation and AI-assisted workforce planning.
• AI-enabled marine and bioeconomy research.
• Research into agentic, neuromorphic and distributed AI systems.
• Software tools for detecting and repairing violations in LLM-generated code.

This pattern suggests that adoption is being driven by specific research problems, available datasets, domain expertise and access to computing resources. It does not yet resemble uniform adoption across the academic workforce.

Health research shows both progress and caution

The University of Auckland’s stroke-recovery research combines MRI scans, image processing and AI models to investigate why patients with apparently similar strokes can have different outcomes. The research uses scans from hundreds of patients and aims to identify patterns that could eventually support more personalised rehabilitation planning. auckland.ac.nz

The project is more than halfway through its doctoral research phase and has produced conference presentations and published research. However, the researchers state that further validation with larger datasets is required, including future work using New Zealand imaging data, before any tool could be used clinically. auckland.ac.nz

This is representative of the current research position: AI is producing useful analytical capability and scientific insight, but many systems remain at the validation or translational-research stage.

Research about AI is expanding alongside research with AI

New Zealand universities are not only applying AI to other disciplines. They are also building research programmes around AI’s social, cultural and technical consequences.

Recent examples include:

• Human-centred and values-led AI research through Huanui.
• Investigation of ethnic and gender bias in LLM-generated clinical vignettes at the University of Canterbury.
• Research into human–robot interaction and perceived robot threat.
• Indigenous data and algorithmic sovereignty research at the University of Waikato.
• Māori sign-language robotics co-designed with Turi Māori, whānau, iwi and hapū.
• Research into agentic AI, distributed systems and brain-inspired computing through international consortia. canterbury.ac.nz

This breadth is an asset. It reduces the risk that the national research agenda becomes limited to imported productivity tools or narrow commercial applications. It also increases the need for shared standards across disciplines with different approaches to evidence, ethics and reproducibility.

Adoption data remains the weakest part of the picture

There is still no published, representative national measure covering:

• Use of AI by academic discipline.
• Differences between researchers, doctoral candidates and professional staff.
• Adoption by career stage, institution or community.
• Use of commercial, open-source or locally hosted models.
• Effects on research productivity, quality or reproducibility.
• Costs, energy use and infrastructure demand.
• The number of AI projects progressing from prototype to sustained use.
• The extent to which Māori and Pacific data or mātauranga Māori are involved in AI projects.

The launch of NZRIS should improve visibility of funded activity, but it will not automatically provide these measures. A separate research-practice survey, combined with workflow and infrastructure data, is still required.

Governance, Policy and Regulation

New Zealand’s national approach remains principles-based

The Government’s AI strategy continues to emphasise adoption and application, with a light-touch regulatory approach based on existing law and OECD principles rather than a standalone AI Act. The strategy explicitly gives greater weight to practical adoption than to competing globally in foundational AI development, while still identifying AI research and advanced technology as national priorities. mbie.govt.nz

For academic research, the practical consequences are that universities, funders, ethics committees, publishers and research teams remain responsible for translating broad principles into operational controls. These controls include:

• Data classification and approved-tool requirements.
• Privacy and security assessments.
• Research-ethics review.
• Intellectual-property and confidentiality controls.
• Disclosure of substantive GenAI use.
• Human verification of outputs.
• Restrictions on entering sensitive or culturally governed information into external systems.

Doctoral governance has moved from advice to formal declaration

The University of Auckland’s GenAI doctoral guidelines took effect on 1 September 2026. Candidates must include a GenAI declaration with their full thesis proposal at confirmation and with their thesis submitted for examination. auckland.ac.nz

The guidelines require candidates to:

• Discuss GenAI use with supervisors.
• Document tools and substantive uses where reasonable.
• Explain how outputs were verified or corrected.
• Disclose content generated or substantively modified by GenAI.
• Accept responsibility for the final work.
• Demonstrate comprehensive understanding of the research during confirmation and examination.

Examiners are instructed not to upload thesis material to external GenAI platforms or use external AI-detection software. This approach prioritises disclosure, research understanding and confidentiality over retrospective detection. auckland.ac.nz

The policy is significant because it applies across the research process, including literature review, data generation, analysis, coding and writing. Its eventual value will depend on implementation consistency and whether the university publishes anonymised evidence about the types and frequency of use.

Sector policy remains more developed around integrity than capability

A September 2026 study of 49 Australian and New Zealand university policies found that academic integrity and ChatGPT were dominant themes, while pedagogy and competency were less prominent. The findings suggest that universities have been more comfortable defining prohibited or risky behaviour than specifying how researchers and students should build durable AI capability. sciencedirect.com

This distinction matters for research. A policy can require disclosure without providing researchers with:

• Approved systems suitable for sensitive data.
• Reproducible prompt and workflow records.
• Training in model evaluation.
• Guidance on AI-assisted coding and statistical analysis.
• Access to local or institutionally controlled models.
• Clear rules for authorship, attribution and intellectual property.

A mature research environment needs both safeguards and capability-building.

Research assessment and funding controls are tightening

Royal Society Te Apārangi’s 2026 guidance prohibits council and panel members from using generative-AI tools to assess proposals because of confidentiality, fabrication and data-retention risks. royalsociety.org.nz

The Society has also designated AI Technologies as a research area for the 2026 Tāwhia te Mana distinguished researcher round. The scope includes machine learning, natural-language processing, computer vision, generative AI, adversarial AI, AI hardware and AI-driven analytics. The scheme retains objectives relating to Māori, Pacific and female research leadership. royalsociety.org.nz

Together, these developments show a dual movement: AI research is receiving more explicit institutional recognition, while AI use in research assessment is subject to tighter confidentiality controls.

Cultural governance is becoming part of technical design

The University of Auckland’s Turi Māori robotics project illustrates a more substantive form of cultural governance. Researchers worked with Turi Māori, whānau, iwi and hapū to develop signs and communication practices before building the AI and robotic system. The project uses computer vision, lip-reading and multimodal modelling, but its design begins with language, identity and community relationships. auckland.ac.nz

The robots are functioning in a research laboratory and are intended to supplement rather than replace human interpreters. There is no evidence of broad deployment in schools, kōhanga reo or health services.

This is a useful model for locally grounded AI research, but it also raises questions that are not answered by technical performance alone:

• Who controls the resulting data and models?
• How are consent and collective interests maintained?
• Can the system be updated without undermining cultural authority?
• What forms of commercialisation are acceptable?
• How are errors or inappropriate outputs escalated?

These questions are increasingly central to academic AI research in Aotearoa.

Case Studies

Case Study 1: NZRIS as research-system infrastructure

Status: Operating public information system; not an AI platform.

NZRIS is now publicly available and provides a common structure for viewing research funding, organisations and reported outputs. Its launch creates an opportunity to map AI-related investment more systematically than was possible through separate funder and university announcements. mbie.govt.nz

Its present contribution is informational rather than computational. NZRIS does not host models, provide research compute or measure researcher behaviour. Nor does the absence of an AI-tagged record prove that no relevant activity exists, because coverage and classification remain incomplete. nzris.govt.nz

The near-term test is whether universities and funders consistently classify AI-related projects and whether the platform expands beyond its initial provider group.

Case Study 2: Horizon Europe AI collaborations

Status: Funded international research projects; not evidence of scaled deployment.

AIMARIA and CoAgent show how New Zealand is using international collaboration to access research scale that would be difficult to reproduce domestically.

AIMARIA connects marine science, biotechnology and AI to develop new approaches to bioprospecting. CoAgent addresses distributed multi-agent generative AI, cloud–edge computing and trustworthy agentic systems. Both projects place New Zealand researchers inside larger consortia and provide postgraduate and research-development opportunities. mbie.govt.nz

The strategic value is not only the funding. Participation can provide access to international datasets, technical standards, research networks and commercial pathways. The risk is that New Zealand partners remain specialised contributors without building durable local infrastructure or intellectual-property ownership.

Case Study 3: Turi Māori robotic sign-language interpreters

Status: Functioning research prototype in a university laboratory.

The University of Auckland project led by Dr Ho Seok Ahn and Dr Tauwehe Tamati is developing robotic interpreters for Turi Māori. The system combines motion capture, facial-expression and lip-reading information with a Weighted Multimodal Network to recognise and communicate Māori sign language through sign, text and speech. auckland.ac.nz

The research began as a Science for Technological Innovation seed project and later received a NZ$1 million Smart Ideas grant to scale the technology over three years. The robots are now built and functioning in the laboratory. auckland.ac.nz

The project demonstrates a form of AI adoption in which cultural co-design is part of the research method rather than a final consultation step. It remains a prototype, and there is no public evidence yet of sustained deployment or independent evaluation in community settings.

Case Study 4: RosterLab’s pathway from PhD research to Health NZ pilot

Status: Commercial system in use with customers; New Zealand Health NZ deployment remains a pilot.

RosterLab originated in University of Auckland doctoral research on hospital rostering and now uses optimisation and AI-assisted workforce-management tools. Its July 2026 HealthX contract covers an initial pilot across four Health New Zealand services. auckland.ac.nz

The earlier Middlemore work suggests meaningful administrative potential, with roster-management time reportedly reduced from approximately 120 hours per month to fewer than 20 hours. However, the current pilot has not yet produced an independently evaluated result across the participating services. auckland.ac.nz

The case illustrates a credible translational pathway:

1. A clearly defined operational problem.
2. Doctoral research and domain collaboration.
3. University commercialisation support.
4. Private investment and company formation.
5. Adoption testing inside a complex public service.

It also illustrates the distance between technical performance and institutional adoption. A system must fit clinical workflows, employment rules, fatigue requirements, staff preferences and accountability structures.

Case Study 5: AI for stroke-recovery prediction

Status: Active doctoral and translational research; not a clinical deployment.

Researchers at the University of Auckland are using MRI data from hundreds of stroke patients to create three-dimensional maps of affected brain pathways and train models that may predict recovery. The research is intended to support more personalised rehabilitation planning. auckland.ac.nz

The project has generated published research and conference presentations, but the team says further validation is required before clinical use. Future research using New Zealand imaging data will be important for testing whether patterns found in international datasets transfer to local populations. auckland.ac.nz

This is a typical high-value research application: AI helps extract patterns from complex data, but clinical legitimacy depends on external validation, representativeness, explainability and safe integration into professional decision-making.

Trends

1. The unit of adoption is becoming the research workflow

The central question is shifting from whether an individual researcher may use a chatbot to whether an entire workflow is appropriately designed.

A mature AI-enabled research workflow needs to specify:

• What data may enter the system.
• Which model or service is approved.
• How outputs are checked.
• What must be disclosed.
• How prompts, versions and outputs are retained.
• How results can be reproduced.
• Who is accountable when the system changes or fails.

Auckland’s doctoral declaration, NZRIS’s structured data model and RosterLab’s domain-specific deployment all point towards this workflow-based approach.

2. International collaboration is compensating for limited domestic scale

The Horizon Europe projects demonstrate a growing strategy: New Zealand researchers contribute specialised expertise to larger international programmes rather than attempting to build every capability domestically.

This is likely to remain important for:

• Foundation-model research.
• Distributed and agentic AI.
• Marine and environmental datasets.
• AI hardware and energy-efficient computing.
• Large-scale health validation.
• International standards and evaluation.

The challenge is ensuring that participation builds local talent, infrastructure, intellectual property and enduring research capability rather than only short-term project activity.

3. Local adaptation is more credible than frontier-scale competition

The most plausible areas of advantage remain those where New Zealand offers distinctive environments, datasets, communities or institutional problems:

• Health and population-specific research.
• Māori and Pacific data governance.
• Agriculture, forestry and aquaculture.
• Marine and environmental science.
• Climate and extreme-weather modelling.
• Remote and autonomous systems.
• Language, culture and accessibility technologies.
• Public-sector and health-system optimisation.

The Turi Māori robotics project and the stroke-imaging research both demonstrate the value of local context. In each case, success depends on whether imported methods can be adapted to local populations, languages and values.

4. Cultural governance is becoming a technical requirement

Māori and Indigenous perspectives are appearing in research methods, project design, university initiatives and funding expectations. The change is most meaningful where communities participate in defining the problem, data practices and acceptable outcomes.

However, public evidence is still stronger for principles and co-design statements than for long-term governance arrangements. The sector has not yet published consistent information on:

• Māori and Pacific decision rights over AI projects.
• Ownership and stewardship of culturally significant datasets.
• Model evaluation against Māori and Pacific populations.
• Commercialisation and benefit-sharing arrangements.
• Long-term maintenance of culturally grounded systems.

5. Research governance is moving faster than impact measurement

Universities are producing guidance, declarations, workshops and approved-tool processes. Funders are introducing AI-related requirements and restrictions. Yet evidence of effects on research quality, speed, reproducibility or equity remains limited.

This creates an asymmetry:

• It is increasingly possible to show that a policy exists.
• It is sometimes possible to show that a project is funded or a prototype is running.
• It is rarely possible to show independently measured sector-wide impact.

The next phase of maturity requires evaluation, not only governance documentation.

6. Announcement activity still exceeds deployment evidence

The sector continues to generate more announcements than independently verified operating results. Stakeholders should distinguish carefully between:

• A research grant and a running system.
• A laboratory prototype and community deployment.
• A pilot and a sustained service.
• A university spin-out and proven public value.
• A vendor-reported efficiency gain and an independently evaluated outcome.
• A national funding commitment and a commissioned platform.

The available evidence supports real and growing adoption in selected research workflows. It does not support claims of uniform or economy-wide academic adoption.

Outlook

Over the next year, the most important indicators will be operational:

• Whether NZIAT names, contracts and launches the national AI research platform or Growth Engine.
• Whether researchers can access shared compute, data, evaluation tools and technical support through that investment.
• Whether NZRIS adds more universities and research organisations and enables reliable AI-related funding analysis.
• Whether AIMARIA and CoAgent produce public research outputs, postgraduate opportunities and durable New Zealand capability.
• Whether Auckland’s doctoral declaration generates usable evidence about actual patterns of GenAI use.
• Whether other universities adopt comparable research-lifecycle controls without creating incompatible requirements for cross-institution teams.
• Whether health-AI projects progress from retrospective datasets to external and prospective validation.
• Whether prototypes such as the Turi Māori sign-language robots move into carefully governed community trials.
• Whether funders and universities publish methods for measuring research quality, reproducibility, environmental cost and productivity effects.
• Whether Māori and Pacific governance is reflected in ongoing decision rights, stewardship and benefit-sharing rather than only in project principles.

A national survey of academic researchers would materially improve policy and institutional decision-making. It should measure not only whether researchers use AI, but which tasks they use it for, what data they provide, how they verify outputs, what barriers they face and whether use differs across disciplines, career stages and communities.

Overall Assessment

As of 1 October 2026, AI in academic research in Aotearoa New Zealand is best characterised as selective, increasingly governed and more internationally connected, but still weakly measured and nationally uncoordinated.

Since the previous update:

• NZRIS has created a new national information layer for publicly funded research.
• Horizon Europe participation has produced new AI research projects with New Zealand partners.
• Huanui has established a visible human-centred AI initiative at the University of Auckland.
• RosterLab has moved university research into a live Health New Zealand pilot.
• Research prototypes in health, robotics and culturally grounded language technology have advanced.
• Doctoral GenAI governance is now operational at a major research-intensive university.

The main uncertainty remains the national AI platform. Government has committed funding and NZIAT has established the institutional framework, but the public record still does not show a named, commissioned and operating national AI research centre.

New Zealand’s academic AI ecosystem is therefore developing through distributed projects, international consortia, university policies, research infrastructure and selected translational pathways. The strategic test is whether those elements become shared, reproducible and independently evaluated capability, with Māori and Pacific governance embedded in how systems are designed and used.