AI in Education

Executive Summary

The direction of travel has become clearer since the previous update, but the underlying adoption gap remains.

Schools continue to show high individual use alongside uneven policy and capability. No newer national survey has displaced the Education Review Office’s July 2026 baseline, which found widespread use by leaders, teachers and students but limited strategic planning.

The most significant new development is the Tertiary Education Commission’s sector-wide procurement of enterprise generative-AI licences and capability-building services. The tender could support access for up to approximately 420,000 staff and students, although that figure is for pricing and evaluation only and does not represent confirmed uptake. The procurement is not evidence of deployment at scale. gets.govt.nz

Other important developments are:

The replacement for NCEA is now substantially more settled. The Government has confirmed a Foundational Award from 2028, the New Zealand Certificate of Education from 2029 and the New Zealand Advanced Certificate of Education from 2030. This supersedes the previous edition’s description of the reform as an open 2026 consultation. education.govt.nz
Applied Intelligent Systems remains a subject under development, with delivery intended for Years 12–13 from 2029. It is not yet an operating school subject. beehive.govt.nz
New Zealand research is beginning to provide evidence for structured use of GenAI in assessment. A Massey University study of 164 first-year students found positive reported effects from a scaffolded assessment design, but did not establish causal improvements in achievement or competence. tandfonline.com
Tertiary institutions are building more visible AI capability through projects such as Unitec’s Ako AI Agents, the AIHE Aotearoa community of practice and the University of Auckland’s Huanui initiative. These are signs of institutionalisation, not proof of sector-wide educational benefit. aihe.nz
The refreshed Years 0–10 curriculum is now being released, including Technology content and the fully updated Te Marautanga o Aotearoa. AI is being positioned as a topic for local curriculum integration and critical inquiry, rather than as a universally mandated standalone subject. newzealandcurriculum.tahurangi.education.govt.nz

The sector is therefore moving from informal use towards managed access, curriculum alignment and assessment redesign. The central unresolved question is whether these developments will improve learning, or mainly make AI more available.

What Has Changed Since the Last Update

Tertiary AI has become a procurement and capability issue

The Tertiary Education Commission opened a request for proposals on 24 September 2026 for enterprise-grade generative-AI services for universities, polytechnics and wānanga.

The proposed arrangements cover:

General-purpose conversational AI.
Advanced reasoning and research.
Document and data analysis.
Multimodal capabilities.
AI assistants and agents.
Administration, identity, security, privacy and usage management.
Implementation support and capability building.

The tender is designed as a syndicated opportunity. One or more suppliers may be selected, but each eligible tertiary education organisation will decide whether to participate and which provider or providers to use. gets.govt.nz

This is a material change from a sector in which institutions largely developed separate policies, licences and pilots. It suggests that government and tertiary leaders now see AI access as shared infrastructure rather than solely a local teaching decision.

It is nevertheless an announcement and procurement process, not an operating deployment. The tender closes on 23 October 2026, and no supplier, contract value or confirmed participation rate has yet been published. The stated 420,000-user figure must not be reported as the number of users who will receive access.

The NCEA replacement is no longer merely provisional

The previous edition described the replacement of NCEA as a consultation due to close on 15 September 2026. That date was incorrect. The relevant public consultation ran from 4 August to 15 September 2025, and the Government reported receiving more than 8,200 submissions when it closed on 16 September 2025. beehive.govt.nz

The Government released further details in May 2026. The confirmed framework is:

A Foundational Award focused on reading, writing and mathematics from 2028.
The New Zealand Certificate of Education for Year 12, beginning in 2029.
The New Zealand Advanced Certificate of Education for Year 13, beginning in 2030.
A staged transition aligned with the refreshed senior-secondary curriculum. education.govt.nz

This matters for AI because assessment conditions, subject structures and authenticity expectations will be rebuilt around a more subject-based system. Applied Intelligent Systems is intended to sit within this emerging framework, but its final curriculum, assessment and qualification arrangements are still being developed.

The reform is therefore both more settled and more consequential than the previous edition suggested. The question is no longer whether the qualification architecture will change, but how AI capability and AI-assisted work will be defined within it.

Curriculum infrastructure is moving from draft to implementation

On 9 September 2026, the Ministry released the Years 0–10 New Zealand Curriculum content for Technology and Learning Languages. On 25 September, the fully updated Te Marautanga o Aotearoa for Years 0–10 became available for schools and kura to begin using in 2027. newzealandcurriculum.tahurangi.education.govt.nz

The curriculum material does not create a single compulsory AI course for all learners. Instead, AI appears through:

Technology learning.
Digital and computational thinking.
Ethical and social inquiry.
Critical evaluation of automated systems.
Local curriculum design.
Discussion of data, bias, sustainability and cultural impacts.

Tāhūrangi also provides an Introduction to Artificial Intelligence resource for kaiako. It explains core concepts, opportunities, risks, Māori data sovereignty and possible classroom activities. The resource explicitly states that it supports local curriculum integration rather than providing operational guidance on individual AI tools. newzealandcurriculum.tahurangi.education.govt.nz

This is a meaningful shift from treating AI only as a disruption to assessment. It positions AI literacy as part of broader technology and citizenship learning. However, the quality and consistency of implementation will depend heavily on teacher capability and local curriculum decisions.

New research supports scaffolded use, but evidence remains limited

A Massey University study published on 12 September 2026 examined a structured GenAI assessment in a fully online first-year health communication course.

The task required 164 students to:

Generate an initial AI-assisted draft.
Critically evaluate its accuracy, bias, theoretical application and cultural implications.
Revise the work in their own words.
Reflect on their use of AI and disclose it transparently.

Students reported positive effects on critique writing, critical thinking and responsible AI use. However, perceptions of reliability were cautious, confidence and competence did not significantly exceed neutral levels, and the study used a cross-sectional survey of student perceptions rather than a controlled comparison of learning outcomes. tandfonline.com

The study strengthens an emerging policy direction: the educational value of AI depends substantially on task design. Requiring students to interrogate and revise AI output is more defensible than allowing unstructured generation for task completion.

It does not show that GenAI improves achievement generally, nor that the approach will transfer unchanged to younger learners, different disciplines or face-to-face settings.

Institutional AI communities are becoming more visible

The 2 September 2026 AI in Higher Education Symposium in Ōtepoti Dunedin has been followed by the development of AIHE Aotearoa as a continuing community of practice for educators, researchers and technologists. The initiative includes recordings, practical experiments and an awards showcase focused on applied projects. aiforum.org.nz

The University of Auckland has also launched Huanui: The AI Initiative. The initiative is intended to support research, teaching, innovation, collaboration and capability-building, with a stated emphasis on human judgement, mātauranga Māori, Indigenous perspectives and values grounded in Aotearoa. auckland.ac.nz

These developments indicate that tertiary AI activity is broadening beyond academic-integrity policies. Institutions are beginning to build research, professional learning, leadership and community functions around AI.

The evidence still does not show how many tertiary providers have institution-wide AI access, how many have approved AI assistants in production, or whether these initiatives are improving learner outcomes.

Current State of AI Adoption

Schools: widespread use remains ahead of system capability

The ERO evidence summarised in the previous edition remains the strongest national picture for schools:

School leaders and teachers report high levels of AI use.
Students use AI for information, ideas, checking and content creation.
Use is concentrated in planning, resource development and administrative work.
School-wide policies and teacher capability are much less consistent.
Assessment and authenticity are driving more formal responses than everyday classroom practice.

Nothing in the September 2026 evidence suggests that this pattern has materially changed. The new curriculum releases and the Applied Intelligent Systems announcement provide future direction, but they do not demonstrate that schools have moved from individual experimentation to coordinated implementation.

The Ministry’s current guidance continues to make teachers responsible for learning and requires human oversight. It states that AI tools for marking should support rather than replace teacher judgement, and that student work submitted for NCEA assessment must be the student’s own. GenAI is not permitted in NCEA external assessment. education.govt.nz

The result remains a two-speed system:

Individual practice is widespread.
Institutional strategy is uneven.
Assessment practice is becoming more deliberate.
Evidence of learning impact is still weak.

Tertiary education: moving towards managed access

Tertiary education now has several overlapping layers of AI adoption:

Local policies covering student use and academic integrity.
Institution-specific AI assistants and agents.
Professional learning and communities of practice.
Research into AI-supported assessment.
Prospective sector-wide licensing and capability procurement.

Unitec and Manukau Institute of Technology’s Ako AI Agents project remains one of the clearest New Zealand examples of an operating institutional deployment. The institutions report more than 50 agents, approximately 3,000 users and more than 8,000 conversations across teaching, learning and student services. These figures are self-reported and do not independently establish educational benefit. livingwhitepaper.com

The TEC procurement could make managed access much more common. Its emphasis on enterprise administration, privacy, identity, security and capability-building is significant: the proposed purchase is not simply for chatbot subscriptions. It is an attempt to create the organisational conditions for sustained use.

However, institution-wide access can also scale poor practice. Licensing does not resolve questions about:

Which assessments permit AI.
What students must disclose.
How staff verify AI-assisted work.
Whether sensitive learner data is permitted in prompts.
How Māori and Pacific data sovereignty is addressed.
How institutions evaluate learning effects.

Assessment: from prohibition towards designed evidence

Assessment remains the main pressure point for AI adoption.

The current direction is not towards unrestricted use or a return to blanket prohibition. It is towards a differentiated model in which:

Some tasks require unaided demonstration.
Some tasks permit AI with disclosure.
Some tasks use AI as an object of critique.
Process evidence, oral explanation, drafts and supervised work become more important.
AI detectors are treated cautiously or rejected as definitive evidence.

The Ministry and NZQA continue to advise that AI tools should not replace teacher or assessor judgement. The 2026 NCEA review and maintenance updates also introduced clearer authenticity guidance across internal assessments. education.govt.nz

NZQA’s automated text scoring remains the clearest production deployment at national scale. The agency has reported use in more than 55,000 writing assessments during 2025, with human checking for responses near the achievement boundary. Budget 2026 allocated $2.1 million for pilots and proofs of concept involving AI and machine learning in marking, moderation and exam development. These broader uses remain development work, not established national deployments. www2.nzqa.govt.nz

The emerging lesson is that AI is most defensible where the assessment task is narrow, the data is controlled, the model is evaluated against human performance and consequential cases receive human review.

Curriculum: AI literacy is broadening beyond computer science

The updated curriculum resources encourage learners to examine AI as a social, technical and ethical system.

The Tāhūrangi materials identify topics including:

How machine-learning and generative-AI systems work.
Bias and misinformation.
Privacy and data sovereignty.
Sustainability and energy use.
The effects of automation on occupations.
Human judgement and responsibility.
Authentic demonstration of foundational skills. newzealandcurriculum.tahurangi.education.govt.nz

This approach is strategically stronger than presenting AI only as a collection of productivity tools. It allows AI to be discussed in technology, social sciences, languages, the arts and vocational learning.

The limitation is delivery capacity. There is no published national measure showing how many schools are teaching these concepts systematically, how much professional learning teachers have received, or whether Māori-medium, Pacific-medium and specialist settings have equivalent resources.

Early learning and equity evidence remain thin

The latest public evidence still concentrates on primary, secondary and tertiary education. There is no comparable national adoption picture for early childhood education.

Evidence is also limited for:

Māori-medium and kaupapa Māori education.
Pacific-medium settings.
Rural and small schools.
Learners with disabilities.
Students using free versus paid tools.
Access to institutionally approved AI.
Long-term effects on achievement, wellbeing and critical thinking.

The risk is not simply unequal access to software. It is unequal access to adult guidance, safe tools, reliable connectivity, culturally appropriate resources and opportunities to practise critical judgement.

Governance, Policy and Regulation

National guidance remains principles-based

The Ministry’s guidance continues to centre:

Teacher responsibility.
Human oversight.
Appropriate and purposeful use.
Privacy and data minimisation.
Authentic student work.
Professional judgement.
Review of school policy as tools change.

This is a useful baseline, but it is not a detailed national operating standard. Schools still need to make many local decisions about approved tools, age restrictions, disclosure, classroom use, data handling and consequences for unauthorised use. education.govt.nz

The gap identified by ERO therefore remains: national principles exist, but they have not yet been translated into consistently applied expectations for every school, learner, teacher and whānau.

Procurement governance is becoming more important

The TEC tender indicates that tertiary institutions are looking for enterprise controls rather than relying on free consumer tools. This is a positive development, provided procurement decisions include educational and cultural criteria as well as technical ones.

For schools, the Safer Technologies for Schools catalogue provides privacy and security assessments of digital products. Its reports cover data collection, security controls, safety features and AI-related risks where relevant. It is an important procurement mechanism, although participation in the catalogue does not itself establish that a product is educationally effective. education.govt.nz

A September 2026 Ministry bulletin also warned schools about a Mathspace data breach affecting more than one million users across Australia and New Zealand. The incident was not presented as an AI incident, but it illustrates a wider point: educational technology adoption carries supply-chain, account-security and data-governance risks independent of model performance. education.govt.nz

Cultural legitimacy is still an implementation test

The Ministry’s guidance acknowledges that mainstream AI models may be weak in te reo Māori, mātauranga Māori, Pacific languages and Polynesian cultures. Tāhūrangi resources also identify Māori data sovereignty as a significant consideration.

This makes cultural assurance more than a question of translation quality. Education providers need to consider:

Whether culturally significant knowledge is being entered into external systems.
Who controls and benefits from data.
Whether outputs misrepresent whakapapa, histories or tikanga.
Whether learners can challenge culturally unsafe outputs.
Whether whānau, hapū and iwi are involved in relevant decisions.

The University of Auckland’s Huanui initiative explicitly places mātauranga Māori and Indigenous perspectives within its AI work. That is a positive institutional signal, but publicly available evidence of comparable practice across schools and tertiary providers remains limited. auckland.ac.nz

Case Studies

Case Study 1: Tertiary Education Commission sector-wide AI procurement

The TEC’s September 2026 RFP is the most important new adoption signal in the tertiary sector.

It combines two elements:

Enterprise licensing and usage capacity.
Implementation and capability building.

The proposed model preserves institutional choice while allowing aggregated purchasing and shared expectations around security, privacy, governance and support. gets.govt.nz

Status: procurement process, not operating deployment.

Why it matters: it could shift tertiary AI from fragmented institutional experimentation towards a more interoperable and professionally supported model.

What remains unknown: supplier selection, participation, cost, approved use cases, data-location arrangements, evaluation requirements and learner outcomes.

Case Study 2: Massey University’s scaffolded GenAI assessment

Massey researchers studied an assessment in which students generated, critiqued and revised AI-assisted content rather than submitting AI output as their finished work.

Among 164 respondents:

83% reported benefits for critical thinking.
82% reported benefits for understanding responsible GenAI use.
71% reported developing critical-appraisal skills.
69% reported developing source-verification skills.
Only 34% regarded GenAI output as reliable.

The findings are encouraging but should be interpreted as student-reported perceptions within one course. The design was not a randomised trial, and the study did not demonstrate improved grades or durable learning beyond the assessment. tandfonline.com

Status: operating course design evaluated through published research.

Why it matters: it provides a concrete model for assessment-by-purpose and shows how disclosure, critique and revision can be built into the task itself.

Case Study 3: NZQA automated text scoring

NZQA’s automated text scoring is the strongest public example of production AI in New Zealand education.

The deployment is bounded to a defined assessment task and supported by human checking near achievement boundaries. NZQA reports faster results and performance comparable with, or better than, human marking in earlier testing. These performance claims are agency-reported rather than independently validated in the published material. www2.nzqa.govt.nz

Status: operating production deployment in a defined assessment context.

Why it matters: it demonstrates the importance of narrow scope, controlled data, human escalation and explicit accountability.

What it does not show: that automated marking is suitable for all subjects, all assessment types or all learners.

Case Study 4: Unitec’s Ako AI Agents

Unitec and Manukau Institute of Technology report more than 50 AI agents supporting approximately 3,000 users across teaching, learning and institutional services.

The agents are designed around specific courses and workflows and are integrated with institutional systems. This is more mature than isolated use of a public chatbot, but the available evidence remains primarily self-reported activity data. livingwhitepaper.com

Status: operating institutional deployment.

Why it matters: it shows that a New Zealand tertiary provider can build a portfolio of purpose-specific agents rather than treating AI as one undifferentiated tool.

Evidence gap: independent measures of learning gain, staff workload, accessibility, retention or equity effects have not been published.

Case Study 5: Applied Intelligent Systems

Applied Intelligent Systems has been approved for development as an industry-led senior-secondary subject. The subject is intended to cover AI-enabled workflows and agents, low- and no-code technologies, machine learning, project work, human oversight and responsible decision-making.

The Government says it is intended for Years 12–13 from 2029 and has allocated part of a $15 million Budget 2026 investment to industry-led subjects. The responsible Industry Skills Board describes the current announcement as the beginning of development and consultation. beehive.govt.nz

Status: curriculum and qualification development.

Why it matters: AI is being treated as a cross-industry capability, not solely an IT specialisation.

Main risk: low-code delivery without sufficient grounding in systems thinking, security, data governance and technical limitations could produce superficial capability. Teacher supply, school access to expertise and delivery arrangements remain unresolved.

Trends

The market is shifting from tools to managed capability

The next phase of adoption is likely to be shaped less by whether educators can access a chatbot and more by whether institutions can provide:

Approved accounts.
Secure identity and access management.
Data boundaries.
Professional learning.
Assessment guidance.
Usage monitoring.
Technical support.
Evaluation.

The TEC tender is the clearest expression of this shift. It also creates a test for government: shared procurement can reduce duplication, but it should not produce a single default platform without meaningful institutional and cultural choice.

Structured use is becoming the preferred pedagogical model

The Massey study, NZQA’s bounded scoring model and the University of Auckland’s assessment approach all point in the same direction: AI should be connected to a defined learning purpose.

The more credible models ask students to:

Verify information.
Compare outputs.
Explain decisions.
Revise AI-generated material.
Reflect on limitations.
Demonstrate independent understanding.

This is different from measuring success by prompt volume, time saved or the number of AI features enabled.

Curriculum and qualifications are moving in parallel

AI literacy is now being developed through three channels:

General curriculum integration in Years 0–10.
A specialist, industry-led Applied Intelligent Systems subject.
AI-aware assessment and qualification design.

This layered approach is sensible in principle. All learners need basic understanding of automated systems, while some need deeper technical and vocational pathways.

The challenge will be coordination. If the general curriculum, industry-led subjects and new national qualifications are designed separately, students may encounter duplicated content, inconsistent terminology and uneven expectations about AI use.

Evidence of activity is still stronger than evidence of impact

New Zealand now has more evidence of:

AI use.
Procurement.
Policy development.
Professional learning.
Agent deployments.
Curriculum planning.
Student perceptions.

It still has much less evidence of:

Improved achievement.
Durable learning.
Reduced teacher workload after verification.
Better outcomes for disabled learners.
Narrower equity gaps.
Improved Māori and Pacific learner outcomes.
Lower assessment fraud.
Long-term effects on critical thinking.

The sector should resist treating user numbers, conversations, licences or positive perceptions as substitutes for outcome evaluation.

Governance remains fragmented

The policy environment has become more substantial, but it is still distributed across:

Ministry of Education guidance.
NZQA requirements.
ERO findings.
Privacy Commissioner guidance.
ST4S assessments.
Institution-level policies.
Vendor terms.
Curriculum resources.
Local school and kura decisions.

The immediate need is not necessarily a single rigid rulebook. It is a coherent framework that clarifies what is nationally consistent and where local decision-making is appropriate.

Outlook

The next six to twelve months will be shaped by four developments.

First, the TEC procurement will move from market engagement towards supplier evaluation. The key question will be whether the final arrangements include meaningful implementation support and evaluation, rather than simply discounted access to commercial models.

Second, the Ministry is expected to continue releasing and refining the Years 11–13 curriculum. Applied Intelligent Systems will need a clear place within the new qualification structure, with assessment that tests both practical capability and critical judgement.

Third, schools will begin implementing the refreshed Years 0–10 curriculum and Te Marautanga o Aotearoa. AI literacy is likely to develop unevenly unless professional learning, exemplars and culturally grounded resources are made readily available.

Fourth, assessment policy will continue to move towards process evidence and purpose-specific AI rules. New qualifications create an opportunity to design these expectations from the beginning rather than retrofitting them after widespread misuse.

Stakeholders should monitor:

The outcome of the TEC AI licensing procurement.
Participation and implementation rates across tertiary providers.
Published evaluation of Unitec’s AI agents.
The final curriculum and assessment design for Applied Intelligent Systems.
School uptake of AI-related curriculum resources.
National data on student and teacher AI use after the ERO baseline.
Independent studies of learning, workload and equity outcomes.
Privacy and security incidents involving education technology providers.
How the new qualifications define independent work, permitted AI assistance and disclosure.

Overall Assessment

AI in New Zealand education is moving from widespread informal use towards institutional infrastructure, but system capability and evidence remain uneven.

The previous edition’s central assessment still holds: adoption is ahead of governance in schools, while tertiary providers are further along in developing managed institutional responses.

The main change is that the tertiary sector now has a plausible route towards coordinated access and capability building through the TEC procurement. At the same time, the curriculum and qualification reforms have made AI more visible as a workforce and learning capability. Applied Intelligent Systems is no longer only a policy idea; it is a defined subject under development for the 2029 senior-secondary environment.

The strongest current examples share four characteristics:

A bounded and clearly defined use case.
Explicit human responsibility.
Assessment or workflow design that requires critical engagement.
Attention to privacy, cultural context and data governance.

The weakest evidence remains concentrated in claims about educational transformation. There is still no robust national evidence that AI has improved learning outcomes across New Zealand, and little published evidence about its effects in early learning, Māori-medium education, Pacific education, disability support or rural schooling.

New Zealand’s priority should therefore be managed capability rather than rapid expansion. The sector needs access to useful tools, but also the professional judgement, cultural assurance, assessment design and evaluation systems required to decide when those tools genuinely improve education.