Latest 1 September 2026

AI in Academic Research in New Zealand: A Living Whitepaper

Aotearoa’s universities are moving from scattered AI experiments to shared principles, auditable doctoral practice and selected research deployments, while the national AI platform remains unawarded and sector-wide adoption data is still missing. (universitiesnz.ac.nz)

Executive Summary

National coordination has advanced, but not yet been completed. The New Zealand Institute for Advanced Technology (NZIAT) officially launched on 21 August 2026 and now has a permanent chair. However, the official AI Research Platform page still does not name a successful phase-two proposal or confirm the release of the up-to-$70 million investment. (beehive.govt.nz)
Universities have adopted a common public position on responsible AI. Universities New Zealand published a sector AI statement on 27 August 2026. It establishes shared principles covering human accountability, research integrity, transparency, privacy, fairness and risk-based governance, while leaving detailed implementation to each autonomous university. (universitiesnz.ac.nz)
Doctoral governance has moved into operation. From 1 September 2026, University of Auckland doctoral candidates must submit a declaration of GenAI use with their full thesis proposal and thesis for examination. This formalises disclosure and auditability at a major research-intensive institution. (auckland.ac.nz)
The strongest evidence of operational AI adoption remains concentrated in selected domains. Health, climate modelling, high-performance computing and tertiary learning support show concrete systems or research workflows. Evidence of routine GenAI use across academic research as a whole remains limited and largely self-reported.
Research funding continues to favour applied and locally adapted AI. New HRC-funded projects include AI for brain-based ADHD assessment and physics-constrained tumour localisation. These are funded research programmes, not evidence of clinical deployment at scale. (hrc.govt.nz)
AI is also becoming a subject of fundamental and interdisciplinary research. A University of Auckland researcher has joined a US$7 million international programme investigating agency, intelligence and meaning in living and artificial systems, with approximately NZ$650,000 supporting Auckland-based postgraduate research. (auckland.ac.nz)
The evidence base still has a major gap: there is no representative, current national measure of how often New Zealand academic researchers use AI, for which tasks, with what productivity effects, or with what distribution across disciplines, career stages and communities.

What Has Changed Since the Last Update

NZIAT has launched, but the AI platform decision remains unresolved

The most important institutional change is that NZIAT is no longer only a proposed vehicle within the science reforms. The Government formally launched it on 21 August 2026 as New Zealand’s fourth public research organisation, with a mandate covering advanced materials, artificial intelligence and quantum technologies. Steve O’Connor was appointed permanent chair on 18 August. (beehive.govt.nz)

This is progress from the previous update, which described NZIAT as being established but still in a foundation phase. It does not, however, resolve the separate AI Research Platform contest. The MBIE page continues to list the five phase-one concepts and the phase-two process without naming a successful platform. The original process anticipated a funding decision in May 2026 and a contract commencing on 1 July 2026, although the published dates were subject to change. (mbie.govt.nz)

The distinction matters. NZIAT now exists institutionally; the national AI research platform has not yet been publicly identified. The delay continues to affect clarity over research leadership, shared infrastructure, talent pipelines and long-term university–industry partnerships.

Universities have moved from individual policies to a sector position

On 27 August, Universities New Zealand published a common Sector AI Position Statement. It recognises that AI is already part of education, research, work and society, while identifying risks including inaccurate information, bias, privacy breaches, intellectual-property misuse, academic misconduct, environmental impacts and loss of trust. (universitiesnz.ac.nz)

The statement is deliberately high-level. It does not replace university policies, and it confirms that each institution will determine its own detailed rules and implementation arrangements. That makes it a coordination mechanism rather than a single sector-wide operating standard.

Its significance is nevertheless substantial. The statement gives universities a common vocabulary around:

Human accountability.
Research and academic integrity.
Disclosure and transparency.
Privacy and confidentiality.
Fairness and inclusion.
Risk-based controls and human review.
Protection of personal, pre-publication, sensitive and culturally significant information.

This is a clear step beyond treating AI as an isolated teaching or assessment issue.

Auckland’s doctoral requirements are now in force

The University of Auckland’s GenAI declaration requirement became operative on 1 September 2026. Candidates must include the declaration with their full thesis proposal at confirmation and with the thesis submitted for examination. (auckland.ac.nz)

The broader doctoral guidance requires candidates to discuss AI use with supervisors, document substantive use, verify outputs, protect sensitive information and disclose relevant use. Examiners are not permitted to upload thesis material to external GenAI tools or use external AI-detection software. (auckland.ac.nz)

The development indicates a shift from general encouragement to research-lifecycle controls. It also reinforces an emerging sector consensus: disclosure and human verification are more defensible than attempting to detect AI use after the fact.

A concrete tertiary deployment provides a stronger adoption signal

The new evidence from Manukau Institute of Technology and Unitec is not a direct measure of academic research adoption, but it is relevant to the operating environment in which research and teaching occur.

The Ako AI Agents project now reports more than 50 agents, approximately 3,000 users and more than 8,000 conversations since its launch in 2025. The agents support nursing, engineering, business, animal sciences, trades, student services, programme development and assessment design. The institution says each agent is tracked through a lifecycle framework covering purpose, data sources, usage, performance and governance. (unitec.ac.nz)

These figures are institution-reported and have not been independently audited. They nevertheless show a transition from isolated demonstrations to a managed portfolio of deployed systems.

Research activity has expanded, but most new projects remain prospective

The HRC’s August Explorer Grant announcements added new AI research in health. University of Otago researchers are developing an AI and EEG-based approach to support more objective cognitive assessment for children with ADHD, with particular attention to adapting international models to small New Zealand samples and testing performance for Māori and Pasifika groups. (hrc.govt.nz)

The University of Auckland also received funding for a physics-constrained AI tool to estimate tumour location during breast surgery. The project will first use existing imaging and surgical data and validate performance in controlled breast phantoms. It is therefore a funded development and validation programme, not a clinical system already operating in hospitals. (auckland.ac.nz)

Current State of AI Adoption

Maturity assessment

Area	Current position
Institutional governance	Moving from institution-specific guidance toward shared sector principles
Doctoral and postgraduate research practice	Becoming more formal, with declarations, supervisor consultation and examiner controls
AI as a research method	Established in selected areas such as health, climate science, spatial analysis and computational biology
Generative AI in everyday research work	Evidently occurring, but national prevalence and impact are not measured
Research computing infrastructure	Established and increasingly capable of supporting machine-learning workloads
Applied and translational research	Active, particularly in health, climate, agriculture and environmental science
AI as a research subject	Expanding across computer science, ethics, public policy, culture, education and fundamental science
Operational deployment inside tertiary institutions	Concrete examples exist, but remain concentrated in a small number of institutions
Commercialisation	Active but concentrated; evidence of scaled university-linked AI firms remains limited
National coordination	Incomplete pending the AI Research Platform decision
Māori, Pacific and cultural governance	Increasingly embedded in principles, funding expectations and research design

AI as a research instrument

Researchers in Aotearoa are using AI and machine learning for:

Prediction and classification.
Image, audio and spatial-data analysis.
Climate and weather modelling.
Health diagnostics and treatment optimisation.
Literature discovery and synthesis.
Coding, workflow automation and data processing.
Simulation and uncertainty analysis.
Translation, editing and research communication.

The clearest operational evidence comes from research computing rather than from general-purpose chatbots. REANNZ reports that its Mahuika high-performance computing platform passed five million jobs in March 2026. Its case studies include machine learning for extreme-heat prediction, localised weather forecasting, brain-tumour research and neurological-condition prediction. (reannz.co.nz)

These examples demonstrate that AI-enabled research is running on national infrastructure. They do not establish that most researchers, or even most research groups, use AI routinely.

AI as an object of research

The research agenda is broadening beyond model development and applied prediction. New Zealand researchers are examining:

Cultural responsiveness and trust.
Māori and Indigenous data sovereignty.
AI literacy and human capability.
AI’s effects on expertise and scholarly communities.
Regulation, ethics and public policy.
AI in healthcare and clinical decision-making.
Agency, autonomy and the nature of intelligence.

The University of Auckland’s new international project on agency and artificial intelligence is an example of fundamental research rather than immediate technology transfer. It will combine mathematical modelling, information theory, philosophy, biology and computer science, while employing postgraduate researchers in New Zealand. (auckland.ac.nz)

This matters strategically. A research ecosystem that only adopts imported tools may gain short-term productivity but remain dependent on external assumptions, platforms and technical priorities. Research about AI helps build the intellectual and institutional capacity needed to evaluate those dependencies.

Infrastructure is more mature than governance measurement

New Zealand has functioning national research-computing infrastructure, research software engineering support and university-based AI capability. REANNZ reports that Mahuika supports data-intensive workloads across genomics, engineering, environmental science and AI applications, while its case studies show researchers receiving help to automate, scale and reproduce machine-learning experiments. (reannz.co.nz)

The less mature area is measurement. Public sources provide project-level descriptions, infrastructure statistics and institutional announcements, but not a consistent national account of:

AI use by discipline.
Adoption by career stage.
Use of commercial versus open models.
Researcher productivity or quality effects.
The number of projects involving Māori or Pacific data.
The cost and environmental footprint of AI-enabled research.
The proportion of AI projects that progress from prototype to sustained use.

As a result, adoption should be described as selective and demonstrable, rather than widespread and quantified.

Governance, Policy and Regulation

A principles-based national environment

The Government’s AI strategy continues to favour adoption of existing AI systems and a light-touch, principles-based regulatory approach rather than a standalone AI Act. Existing privacy, consumer-protection, human-rights and intellectual-property frameworks are expected to apply, with further intervention if new risks or legislative gaps emerge. (mbie.govt.nz)

For academic research, this places greater responsibility on universities, funders, ethics committees and publishers. The practical controls are therefore likely to be institutional and contractual:

Data-classification rules.
Approved-tool lists.
Research-ethics requirements.
Disclosure obligations.
Publisher and funder policies.
Intellectual-property agreements.
Cybersecurity and procurement assessments.

The Ministry for Regulation’s guidance to regulators similarly emphasises starting small, applying safeguards and retaining human responsibility for decisions. Although written for regulators, the approach is consistent with how universities are beginning to govern higher-risk AI uses. (regulation.govt.nz)

Sector-wide principles, institution-specific implementation

The Universities New Zealand statement provides a common baseline but expressly preserves institutional autonomy. This creates a practical balance:

Common principles make sector expectations more legible.
Institutional policies can reflect disciplinary and cultural contexts.
Cross-institution research teams may still face different disclosure, procurement and data-handling requirements.
Researchers moving between universities may need to maintain multiple compliance pathways.

The University of Otago illustrates this flexible model. Its postgraduate research advice chose not to become a formal policy because GenAI use varies substantially across disciplines and the technology is changing quickly. The advice instead emphasises responsible use, consultation, transparency, declaration and examiner guidance. (otago.ac.nz)

Otago’s staff-use policy is more operational. It requires staff to use approved systems for different data classifications, notify cybersecurity teams before deploying new AI systems, designate system owners and obtain additional approval for some Māori or Pacific data uses. (otago.ac.nz)

Research integrity and data sovereignty

The Royal Society Te Apārangi’s national guidelines remain an important reference point. They frame GenAI use around research integrity, transparency, privacy, intellectual property, environmental impact and the protection of Māori and Pacific data sovereignty. (royalsociety.org.nz)

The key governance challenge is not simply whether a researcher uses AI. It is whether the tool is appropriate for the data and research purpose. Risks are particularly acute where systems process:

Personal or health information.
Unpublished findings.
Commercially sensitive material.
Research-participant data.
Mātauranga Māori.
Te reo Māori resources.
Culturally significant or collectively governed information.

A general-purpose external model may offer convenience while weakening control over storage, reuse, model training and jurisdiction. University policies are increasingly responding through data classification and approved-tool processes rather than blanket permission or prohibition.

Research funding and assessment

AI is now explicitly recognised within the 2026 Tāwhia te Mana Distinguished Researcher Fellowship round. The eligible field includes machine learning, natural-language processing, computer vision, generative AI, adversarial AI, AI algorithms, AI-specific hardware and AI-driven analytics. The scheme retains objectives relating to Māori, Pacific and female research leadership. (royalsociety.org.nz)

Research assessment is also becoming more cautious. Royal Society fellowship guidance warns against using GenAI to assess proposals because of confidentiality risks, fabricated content and the possibility that submitted material could be stored or reused by external systems. (royalsociety.org.nz)

Case Studies

Case Study 1: NZIAT and the unresolved national platform

NZIAT’s launch is a substantial structural development. The organisation is intended to connect advanced-technology research with commercialisation, international partnerships and high-value talent. It has already committed funding to the Future Magnetic and Materials Technologies Platform and is progressing AI and quantum work. (beehive.govt.nz)

The AI component remains incomplete. Five concepts received seed funding for full proposals, covering agentic AI, creative AI, autonomous systems, bioeconomy AI and physical AI for real-world environments. The public MBIE material still does not identify the final platform recipient. (mbie.govt.nz)

This is not evidence that the initiative has failed. It is evidence that the institutional launch and the substantive platform award are separate milestones, and only the first is now confirmed.

Case Study 2: REANNZ and machine learning for climate extremes

University of Auckland researcher Dr Emily Gordon used REANNZ support to automate thousands of machine-learning training runs for research into extreme heat events. The workflow used Nextflow to manage preprocessing, training and postprocessing across high-performance computing resources.

According to REANNZ, the workflow reduced manual intervention, improved reproducibility and enabled broader exploration of model uncertainty. The organisation reports that neural networks outperformed logistic regression in the project’s prediction experiments. These are project-level research findings reported by the infrastructure provider, not an independent evaluation of general AI productivity across the sector. (reannz.co.nz)

The case illustrates a practical maturity pathway: the value came not only from choosing a model, but from integrating machine learning with reproducible workflow engineering and national compute.

Case Study 3: AI and locally adapted health research

The HRC-funded ADHD project illustrates both the promise and the limits of New Zealand’s research position. The team plans to adapt models trained on international datasets using smaller local samples, including Māori and Pasifika participants. The aim is to improve the relevance of brain-based cognitive assessment for New Zealand children. (hrc.govt.nz)

This is strategically important because New Zealand is unlikely to match the largest international datasets. Local value is more likely to come from careful adaptation, validation, cultural responsiveness and implementation in the health system.

The project remains at the research and development stage. It should not be described as an AI diagnostic tool already deployed in clinical practice.

Case Study 4: Ako AI Agents at MIT and Unitec

Ako AI Agents provides one of the clearest publicly described examples of sustained AI deployment in a tertiary environment. More than 50 agents support approximately 3,000 users and have handled more than 8,000 conversations since 2025. The system is integrated into the learning-management environment and includes course-specific tutoring, skills practice, feedback and an auditable nursing assessment simulator. (unitec.ac.nz)

The project’s governance model includes visibility over each agent’s purpose, data sources, usage patterns, performance and lifecycle. This is relevant to academic research because it demonstrates a model for treating AI systems as managed institutional assets rather than informal tools.

The figures are self-reported by MIT and Unitec, and the project concerns teaching and learner support rather than scholarly research. It is therefore best used as evidence of broader tertiary-sector capability, not as a proxy for national research adoption.

Trends

1. The unit of adoption is becoming the governed workflow

The sector is moving beyond the question of whether researchers may use ChatGPT or another tool. The more consequential questions are:

What data enters the system?
What model or service is approved?
How are outputs checked?
What must be disclosed?
Who owns the resulting work?
How is the workflow reproduced?
When must a human intervene?
What happens when the tool changes?

Auckland’s doctoral declaration, Otago’s data-classification requirements and REANNZ’s workflow engineering all point in the same direction: adoption is becoming a matter of process design and accountability.

2. Local adaptation is a stronger strategy than frontier-scale competition

New Zealand’s most credible opportunities remain areas where local context provides an advantage:

Health systems and population-specific research.
Climate and weather extremes.
Agriculture, aquaculture and forestry.
Environmental monitoring.
Māori and Pacific data governance.
Spatial and geophysical data.
Creative and cultural production.
Autonomous systems operating in difficult outdoor environments.

The HRC ADHD project demonstrates the value proposition clearly: adapting international models to local populations may be more realistic and socially valuable than attempting to build the largest models from scratch. (hrc.govt.nz)

3. Cultural responsiveness is becoming an operating requirement

Cultural considerations are no longer confined to general statements about ethics. They are appearing in research design, funding expectations, university policies and system-development practices.

The Universities New Zealand statement identifies culturally significant information as requiring protection. Otago’s policy refers directly to Te Ao Māori principles, Māori data sovereignty and Pacific data sovereignty. The HRC ADHD project also treats representation and local adaptation as technical requirements rather than optional add-ons. (universitiesnz.ac.nz)

4. AI research is becoming more interdisciplinary

The emerging portfolio spans computer science, medicine, climate science, education, philosophy, public policy, cultural studies and information infrastructure. The University of Auckland’s agency research programme is a particularly clear example of AI-related research extending into fundamental questions about life, intelligence and autonomy. (auckland.ac.nz)

This breadth is an advantage for Aotearoa, but it also creates coordination challenges. Researchers may operate under different standards for evidence, reproducibility, ethics and disclosure depending on discipline and funding source.

5. Deployment evidence remains more limited than announcement activity

There is a persistent distinction between:

A funding award and a running system.
A pilot and a sustained service.
A prototype and a validated research instrument.
A university announcement and independently measured impact.

The current evidence supports real adoption in specific workflows and institutions. It does not support claims that AI has been adopted uniformly across New Zealand academic research.

6. National coordination is lagging behind institutional practice

Universities, REANNZ, funders and research teams are building practical capability while the national AI platform remains unresolved. This creates a paradox: bottom-up adoption is becoming more concrete at the same time as the proposed top-down coordination mechanism remains uncertain.

If the platform is confirmed, it could consolidate existing activity. If uncertainty continues, universities and research organisations are likely to keep developing distributed capability, with possible duplication in infrastructure, policy and talent recruitment.

Outlook

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

Whether NZIAT announces and contracts the national AI Research Platform.
Whether the platform funds shared compute, data, evaluation and talent mechanisms rather than only individual projects.
Whether universities align their detailed policies sufficiently to support cross-institution research.
Whether Auckland’s doctoral declaration produces usable evidence about patterns of AI use.
Whether HRC-funded health projects progress from model development to external validation and prospective evaluation.
Whether REANNZ publishes more evidence on AI workload demand, reproducibility and researcher outcomes.
Whether Māori and Pacific governance is reflected in decision rights, consent and data stewardship rather than only in principles.
Whether funders and institutions develop methods to assess the environmental and financial costs of AI-enabled research.

A national adoption survey would materially improve decision-making. Without one, policymakers and university leaders will continue to infer sector maturity from a mixture of case studies, policy documents, infrastructure statistics and grant announcements.

Overall Assessment

As of 1 September 2026, AI in academic research in Aotearoa New Zealand is best characterised as selective, increasingly governed and strongly application-oriented, but still weakly measured and nationally uncoordinated.

The sector has moved forward since the previous update:

NZIAT has formally launched.
Universities now have a shared public position on responsible AI.
Auckland’s doctoral disclosure requirement is operational.
Tertiary institutions are reporting managed deployments of AI agents.
Research funding continues to support locally adapted AI in health and other priority areas.
Fundamental and interdisciplinary research on AI itself is expanding.

The central uncertainty remains the national AI Research Platform. Its absence from the public record is now more consequential because the wider institutional environment has moved on: governance systems are being implemented, research workflows are scaling and universities are making decisions about capability without a confirmed national centre of gravity.

The direction of travel is clear. AI is becoming part of New Zealand’s academic research infrastructure, methodology, governance and intellectual agenda. The next test is whether the sector can convert this distributed activity into reproducible, culturally grounded and independently evidenced research capability, rather than allowing adoption to remain a collection of promising but uneven projects.

← Back to all articles View as Markdown ⚑ Flag an issue