AI in Education in New Zealand: A Living Whitepaper

AI is now routine in New Zealand education workflows, but strategic, system-wide adoption remains limited. New ERO evidence shows high use, uneven rules, capability gaps and growing pressure to redesign assessment around authentic learning.

Executive Summary

New Zealand education has moved beyond isolated experimentation with AI, but it has not yet reached consistent, system-wide implementation.

The strongest new evidence is the Education Review Office’s national review, published on 30 July 2026. It found that:

• 93% of school leaders use AI.
• More than four in five teachers use AI, mainly for planning and resource development.
• Three in four students use AI for schoolwork.
• Around three in four schools remain in an “unplanned” stage of AI adoption.
• Only 28% of leaders report that their school has guidance for teachers using AI.
• Only 26% report a policy covering student use of AI for learning.
• More than half of leaders say they do not know enough to use AI properly. (evidence.ero.govt.nz)

The previous edition correctly identified a shift towards curriculum-linked tools, formal professional learning, and more deliberate assessment policy. That direction remains valid. However, the latest evidence shows that formal announcements and high individual usage are running ahead of institutional capability and governance.

The most consequential developments are:

• ERO has called for a national AI-in-education framework and nationally consistent guidance.
• Applied Intelligent Systems has been approved for development as a senior-secondary subject, but it is not yet an operating course.
• Education Perfect’s Ministry-funded Year 9–10 mathematics platform is available to schools and includes an AI-powered function, although participation is optional and national uptake has not been published.
• NZQA’s automated text scoring is the clearest production deployment, while wider AI use in NCEA remains at the pilot and proof-of-concept stage.
• Unitec and Manukau Institute of Technology report a more mature tertiary deployment, with more than 50 AI agents supporting approximately 3,000 users.
• The proposed replacement of NCEA introduces further uncertainty about the final qualification and assessment settings into which AI-related subjects and practices will fit. (tahurangi.education.govt.nz)

The overall picture is therefore one of high exposure, growing structured experimentation, and incomplete system readiness.

What Has Changed Since the Last Update

ERO has supplied the missing national school-level evidence

The most important addition to the evidence base is ERO’s Ready or not: How are schools responding to Artificial Intelligence? review. Although the report was published on 30 July, it was not reflected in the previous 19 August edition. It draws on surveys, school visits, interviews and focus groups conducted during Term 1 of 2026. (evidence.ero.govt.nz)

The report materially strengthens, and partly qualifies, the previous article’s conclusions:

• AI use is more widespread among school leaders and teachers than earlier survey evidence suggested.
• School-level policy and capability are weaker and more variable than usage rates imply.
• Student use is common, including use that teachers do not always see.
• The benefits are most credible when AI supports planning, idea generation, problem-solving or checking work, rather than completing assessed work.
• Teachers report workload benefits, but secondary teachers also face increased work authenticating student submissions. (evidence.ero.govt.nz)

ERO’s findings shift the central question from whether schools are using AI to whether they are using it deliberately, safely and consistently.

The NCEA transition has become a strategic uncertainty

The previous edition linked Applied Intelligent Systems to the planned NZCE and NZACE senior-secondary qualifications. Since then, the Government has opened consultation on a proposal to replace NCEA with a new qualification pathway.

The proposal would remove Level 1, introduce a new foundational award, move towards a more structured subject approach, and give industry skills boards a greater role in vocational pathways. Consultation is open from 4 August to 15 September 2026. (ncea.education.govt.nz)

This does not cancel the approved Applied Intelligent Systems subject. It does mean that the subject’s final qualification packaging, assessment model and implementation context should now be treated as provisional until the wider qualification reforms are settled. The subject remains an approved development initiative, not evidence of current classroom adoption. (tahurangi.education.govt.nz)

Tertiary AI has moved from discussion towards operating systems

The previous edition highlighted policies, professional networks and planned events in higher education. A new Unitec announcement provides stronger evidence of an operating institutional deployment.

MIT and Unitec report that their Ako AI Agents project has developed more than 50 AI agents since its launch in 2025. The agents support approximately 3,000 users and have been used in more than 8,000 conversations across nursing, engineering, business, animal sciences, trades, student services, programme development and assessment design. (unitec.ac.nz)

These figures are institutionally self-reported and do not independently establish learning gains. They do, however, demonstrate a transition from one-off classroom experiments to a managed portfolio of AI services integrated with teaching and institutional workflows.

Professional learning has become a more formal market

The University of Canterbury’s UC Online has expanded from short courses into a postgraduate certificate in Artificial Intelligence and Digital Education. The programme begins with a September 2026 intake and covers AI literacy, authentic assessment, AI-enabled learning design, cultural responsibility, data governance, equity and learner agency. (uconline.ac.nz)

The programme is evidence of growing demand for formal capability development, but enrolment numbers and participant outcomes have not been published. It should therefore be understood as an indicator of sector infrastructure, not as evidence that educator capability has already improved nationally.

National coordination is visible, but the national framework is still missing

The AI Forum’s 2026 Blueprint continues to promote a cross-curricular approach to AI literacy, equitable access to AI tools and professional development. It also calls for a five-year strategy and a framework mapping AI capabilities across age groups and workforce roles. (aiforum.org.nz)

ERO’s recommendation for a national AI-in-education framework confirms that this work remains incomplete. New Zealand has principles-based Ministry guidance, NZQA assessment guidance and sector networks, but not yet a single nationally adopted framework covering curriculum, teacher use, student use, assessment, procurement, privacy and whānau communication. (evidence.ero.govt.nz)

Current State of AI Adoption

Schools: high use, low consistency

AI is now a routine productivity tool for many school leaders and teachers.

ERO found that 93% of school leaders use AI. The most common uses are:

• Drafting communications: 80%.
• Summarising information: 70%.
• Creating learning resources: 66%.
• Generating school policies or guidance: approximately half.
• Curriculum planning: 52% of primary leaders and 35% of secondary leaders. (evidence.ero.govt.nz)

More than four in five teachers use AI. Their most common applications are tailoring teaching resources, at 79%, and lesson planning, at 65%. Use for marking and feedback is lower, at 22%, reflecting concern about reliability, fairness and the responsibility attached to assessing student work. (evidence.ero.govt.nz)

These findings are consistent with TALIS 2024, although the surveys measure different populations and were conducted at different times. TALIS found that 69% of New Zealand Year 7–10 teachers had used AI in the previous year, compared with 36% across the OECD. Among New Zealand teachers who had used AI, 78% had used it to generate lesson plans or activities and 73% to learn about or summarise a topic. Only 12% had used it to assess student work. (educationcounts.govt.nz)

The evidence supports a clear distinction:

AI adoption is already mainstream in teacher preparation and administrative work, but much less mature in assessment, student-performance analysis and institution-wide learning design.

School governance has not kept pace with use

ERO classifies around three in four schools as taking an “unplanned” rather than strategic approach to AI. Only 28% of leaders report that their school has guidance for teachers using AI. Only 26% report a policy covering student use of AI for learning. Around 25% say individual teachers set their own rules for students. (evidence.ero.govt.nz)

Secondary schools are further ahead in assessment governance:

• 77% of secondary leaders report having a policy for student use of AI in assessment.
• No secondary leaders in the ERO sample reported having no assessment policy at all.
• However, only 43% of secondary students who knew their school had rules considered those rules useful. (evidence.ero.govt.nz)

The result is a two-level system. Assessment requirements have prompted more formal responses in secondary schools, while everyday classroom use, teacher practice and primary-school guidance remain more dependent on local decisions.

Students are active users, often without formal instruction

ERO found that three in four students use AI for schoolwork. The most common uses are finding information and generating ideas:

• 72% of secondary students and 62% of primary students use AI to find information.
• 63% of secondary students and 54% of primary students use it to generate ideas.
• 66% of secondary students use AI to check work, compared with 45% of primary students.
• 50% of primary students report using AI to generate music or images, compared with 31% of secondary students. (evidence.ero.govt.nz)

Students are also largely teaching themselves. This is reported by 77% of secondary students and 58% of primary students. Nearly four in ten parents and whānau say they are unsure whether their children use AI at school. (evidence.ero.govt.nz)

The gap between student use and formal instruction creates a risk of invisible practice. Students may understand how to operate tools without understanding accuracy, bias, privacy, age restrictions, attribution or the difference between assistance and substitution.

Impacts on workload and learning are mixed

More than half of teachers report that AI makes their job easier, while only 7% say it makes their job harder. The reported time savings are concentrated in resource creation and planning:

• 61% of teachers report spending less time creating tailored teaching resources.
• 52% report spending less time on lesson planning.
• 55% report that AI makes their job easier overall.
• 90% of leaders believe AI makes it easier for teachers to do their jobs. (evidence.ero.govt.nz)

The workload effect is not uniformly positive. Secondary teachers who encounter student misuse are 5.5 times more likely to report increased time spent authenticating student work than teachers who do not encounter misuse. (evidence.ero.govt.nz)

Perceived effects on students are also divided. Fifty-seven percent of teachers believe AI is worsening critical thinking, compared with 15% who believe it is improving critical thinking. Teachers are more likely to report positive effects when AI is used for idea generation, problem-solving or checking work, rather than producing the work itself. These are reported perceptions, not controlled measures of learning outcomes. (evidence.ero.govt.nz)

Tertiary education: institutional responses are more developed

Higher education is increasingly moving from general statements about academic integrity to specific operating models.

The University of Auckland’s two-lane approach distinguishes between:

• Lane 1: controlled assessments where AI may be restricted.
• Lane 2: other assessments where AI may be used, with students remaining responsible for their submissions.

The University has embedded the approach in assessment procedures and is using dedicated staff resources and professional learning to support implementation. (teachwell.auckland.ac.nz)

Unitec’s Ako AI Agents project provides a more operational example. Its agents are integrated into the learning management system and provide course-specific tutoring, skills practice and feedback. The project uses a full-lifecycle framework for development, evaluation, maintenance and governance, with kaiako and ākonga involved in identifying and designing use cases. The reported scale is significant for a New Zealand tertiary provider, but its educational impact remains to be independently evaluated. (unitec.ac.nz)

There is no published national benchmark showing how many tertiary institutions have approved AI tools, active agents, AI-related policies or student AI-literacy programmes. The available evidence is therefore institution-specific and should not be generalised to the whole tertiary sector.

National assessment: the clearest production use remains NZQA

NZQA’s automated text scoring remains the most mature public-sector education deployment.

NZQA reports that more than 55,000 writing assessments were marked using automated text scoring in 2025. Responses near the achievement boundary were check-marked by experienced human markers, and results were returned approximately 3.5 weeks earlier than the previous year. NZQA also reports that its earlier pilot produced results comparable to, or more accurate than, human marking. These performance claims are agency-reported rather than independently validated in the published material. (www2.nzqa.govt.nz)

Budget 2026 provided $2.1 million for NZQA to develop pilots and proofs of concept for AI and machine learning in marking, moderation and exam development. The stated uses include extending AI marking to selected end-of-year NCEA assessments, supporting moderation and assisting exam-development teams. This funding is evidence of planned expansion, not evidence that those broader uses are already operating at scale. (education.govt.nz)

The key lesson from NZQA is not that automated marking can replace professional judgement. It is that limited AI use can operate credibly where the use case is narrow, the data is controlled, boundary cases receive human review, and accountability remains explicit.

Early learning and system-wide equity evidence remain thin

The most recent national AI evidence concentrates on schools, particularly primary and secondary settings. The ERO review does not provide a comparable national adoption picture for early childhood education, and there is no equivalent published national benchmark for tertiary education.

Evidence is also limited on:

• Māori-medium and kaupapa Māori settings.
• Pacific-medium education.
• Rural and small-school implementation.
• Learners with disabilities and learning support needs.
• Differences in access to paid or institutionally approved tools.
• Measurable effects on achievement, engagement or wellbeing.

ERO found that smaller schools may struggle to sustain an “AI champion” model because of limited staffing and the risk of depending on one individual. This suggests that capacity constraints may become more important than simple access to software. (evidence.ero.govt.nz)

Governance, Policy and Regulation

Ministry and NZQA guidance

The Ministry’s current guidance is principles-based. It places teachers and teacher–student relationships at the centre, requires human oversight and states that AI should support rather than replace professional judgement. It advises schools to create policies covering purpose, acceptable use, risk mitigation, professional development, data privacy and review. (education.govt.nz)

For assessment, the Ministry states that:

• Evidence submitted for assessment must be the student’s own work.
• GenAI use is not permitted in NCEA external assessment.
• Internal assessment use depends on the relevant achievement standard and school policy.
• Teachers should not rely on AI tools to make final marking decisions.
• Schools with consent to assess standards must have an authenticity policy that includes acceptable GenAI use. (education.govt.nz)

NZQA’s tertiary guidance similarly encourages coherent assessment and moderation systems, clear communication with learners, a range of assessment methods and regular review of academic-integrity policies. (www2.nzqa.govt.nz)

ERO’s call for a national framework

ERO recommends nine actions under three broad areas:

• A national AI-in-education framework adapted to the New Zealand context.
• Nationally consistent guidance for teachers, students, assessment and parents or whānau.
• Support for curriculum integration, AI literacy, capability building and equitable implementation. (evidence.ero.govt.nz)

The report’s recommendation is significant because it identifies a structural limitation in the current model. Schools have local flexibility, but the absence of common expectations produces inconsistent rules for students and staff. ERO found that approximately half of leaders and teachers lack clear guidance on AI use, while around half do not consider current AI tools trustworthy or reliable. (evidence.ero.govt.nz)

Privacy and children’s data

The Office of the Privacy Commissioner’s education-sector guidance, released in March 2026, treats AI as part of the wider digital-technology environment governed by the Privacy Act 2020. Education providers remain responsible for how learner information is collected, used, stored and shared, even where a third-party vendor operates the technology. (privacy.org.nz)

The guidance warns that providers should not rely solely on vendor assurances. They should undertake their own due diligence, consider a Privacy Impact Assessment, understand whether learner information is used to train models, and maintain records of approved technologies. It also highlights risks from automated profiling, inferred learner characteristics, inaccurate or biased outputs, and chatbots that encourage children to disclose personal information. (privacy.org.nz)

The Privacy Commissioner specifically recommends examining whether:

• A tool collects more information than it needs.
• Data is retained after a learner leaves.
• Information is shared with third parties.
• New AI functionality has changed an existing product’s risk profile.
• The provider has adequate access controls, staff training and deletion processes.
• Learners and parents or whānau understand how information is being used. (privacy.org.nz)

Cultural legitimacy and data sovereignty

The Ministry cautions that many AI models are weaker in mātauranga Māori, te reo Māori, Pacific languages and Polynesian cultures. This is not only an accuracy issue. It affects whether a system is culturally safe, whether its outputs can be trusted, and whether locally significant knowledge is being processed appropriately. (education.govt.nz)

University policies and emerging professional-learning programmes increasingly refer to Te Tiriti o Waitangi, Māori and Pacific data sovereignty, cultural responsiveness and equity. However, published evidence of how these principles are being applied in operating school and tertiary systems remains limited.

Case Studies

Case Study 1: Unitec’s Ako AI Agents

Ako AI Agents is currently the strongest publicly reported example of institution-wide tertiary AI deployment in New Zealand.

Since launching in 2025, the project reports:

• More than 50 AI agents.
• Approximately 3,000 users.
• More than 8,000 conversations.
• Deployments across nursing, engineering, business, animal sciences, trades and student services.
• Use in programme development and assessment design. (unitec.ac.nz)

The project’s significance lies in its operating model. Agents are integrated into the learning management system and designed around specific teaching, learning or service needs rather than offered as unrestricted general-purpose chatbots.

The limitations are equally important. The available announcement is self-reported, does not provide independent outcome evaluation, and does not establish whether the reported interactions improved achievement, reduced staff workload or changed student retention.

Case Study 2: NZQA automated text scoring

NZQA’s writing assessment deployment illustrates a bounded public-sector use case.

The system is used for high-volume writing assessments in a controlled environment. NZQA reports faster results and maintains human checking for responses near achievement boundaries. It is now using Budget 2026 funding to investigate further applications in marking, moderation and exam development. (www2.nzqa.govt.nz)

This is a production deployment in one defined assessment context, not a general endorsement of automated marking across schools. The model depends on:

• A narrowly specified task.
• Secure assessment data.
• Evaluation against human marking.
• Human review of difficult or consequential cases.
• Clear responsibility for final outcomes.

Case Study 3: Ministry-funded Education Perfect mathematics platform

The Ministry-funded Education Perfect platform gives Year 9 and 10 students and teachers access to digital mathematics resources from Term 1 2026. The platform includes structured lessons, assessment tools, learner-progress insights, inclusive features and an AI-powered tool. Schools may opt in, and Education Perfect is providing professional learning. (tahurangi.education.govt.nz)

Education Perfect describes additional AI-supported functions including reading assistance, hints, learning loops and feedback on extended responses. These details are provider-reported and should not be interpreted as independent evidence of learning gains. (educationperfect.com)

The deployment is important because it represents a shift from teacher-by-teacher experimentation to AI functionality supplied through a curriculum-linked platform. However, the Ministry has not published national participation rates, usage volumes, privacy-assessment results or outcome data.

Case Study 4: Applied Intelligent Systems

Applied Intelligent Systems has been approved as one of nine industry-led senior-secondary subjects. The planned subject will address artificial intelligence, machine learning, agents, large language models, workflow automation, failure modes, bias and human oversight. It is intended to combine practical project work with critical evaluation and real-world problem-solving. (tahurangi.education.govt.nz)

This is a significant curriculum signal: AI is being positioned not only as a tool used in other subjects, but also as a workforce-relevant area of study.

It is not yet an operating deployment. The subject is still being developed, and its eventual assessment and qualification arrangements may be affected by the current consultation on replacing NCEA.

Case Study 5: University-level assessment and capability building

The University of Auckland has embedded a two-lane approach that permits AI in some assessment settings while preserving controlled assessments where independent performance must be demonstrated. The model represents a move away from blanket prohibition towards assessment-by-purpose. (teachwell.auckland.ac.nz)

The University of Canterbury is complementing short courses with a postgraduate certificate that prepares educators to evaluate AI tools, redesign authentic assessment and develop culturally responsive AI-enabled learning. The course is an education and capability intervention rather than evidence of a live AI system. (uconline.ac.nz)

Trends

AI adoption is moving through bounded institutional environments

The strongest examples now connect AI to a defined subject, course, workflow or assessment process:

• Education Perfect links AI functionality to mathematics resources.
• NZQA links automated scoring to a specific writing assessment.
• Unitec links agents to courses and institutional services.
• The University of Auckland links permitted use to assessment design.

This bounded approach improves accountability and makes it easier to define data boundaries, acceptable use, human responsibilities and evaluation criteria.

Workflow adoption is ahead of pedagogical transformation

Most current use remains concentrated in drafting, summarising, lesson planning and resource creation. These uses may reduce workload, but they do not necessarily change how students learn.

The more difficult work involves:

• Designing learning activities that build judgement rather than dependence.
• Teaching students to test AI outputs.
• Integrating AI literacy across subjects.
• Evaluating whether AI improves learning, not just speed.
• Maintaining authentic evidence of student capability.

ERO’s finding that only 31% of leaders explicitly teach students about AI, while 68% report informal teaching through classroom activities, shows that AI literacy is still more incidental than systematic. (evidence.ero.govt.nz)

Assessment redesign is replacing detector-led strategies

The current direction is towards controlled assessments, oral explanations, drafts, learning journals, process evidence and explicit disclosure of AI use.

ERO reports that some secondary schools are increasing supervised assessment, reviewing drafts and planning notes, and asking students to explain their learning processes. These approaches are more labour-intensive but better aligned with the goal of demonstrating learning than relying solely on unreliable AI detectors. (evidence.ero.govt.nz)

The NCEA transition makes this issue more urgent. Any future qualification system will need to distinguish between:

• Skills students must demonstrate independently.
• Professional contexts in which AI use is expected.
• Assessment tasks where AI is permitted but must be disclosed.
• Tasks where AI would invalidate the evidence.

Capability is becoming the main adoption constraint

The most important divide is no longer simply whether educators have access to AI. It is whether they can evaluate and govern it.

ERO found that:

• 56% of leaders do not know enough to use AI properly.
• 45% of teachers report the same limitation.
• 53% of leaders and 48% of teachers lack clear guidance.
• 48% of leaders and 53% of teachers do not consider current AI tools trustworthy or reliable. (evidence.ero.govt.nz)

This suggests that professional learning must include pedagogy, assessment, privacy, cultural competence, procurement and critical evaluation—not merely prompting techniques.

The policy gap is becoming more visible

New Zealand has substantial activity across government, universities, professional networks and vendors, but the system remains fragmented.

The country now has:

• Ministry school guidance.
• NZQA assessment guidance.
• Privacy Commissioner guidance.
• ERO recommendations.
• AI Forum coordination.
• University policies and professional-learning programmes.
• Local school and kura policies.

What it does not yet have is a nationally adopted framework that joins these elements together and translates them into practical, age-specific expectations.

Evidence of learning impact remains weak

The available New Zealand evidence measures use, perceptions, policy and workload more effectively than learning outcomes.

There is still limited independent evidence on whether AI:

• Improves achievement.
• Reduces teacher workload after verification and oversight are included.
• Narrows or widens equity gaps.
• Improves accessibility for learners with additional needs.
• Strengthens or weakens critical thinking over time.
• Produces durable gains beyond the immediate task.

Vendor and institution-reported activity metrics are useful indicators of adoption, but they should not be confused with evidence of educational effectiveness.

Outlook

The next phase will be shaped by three connected processes.

First, the Government’s NCEA consultation, which closes on 15 September 2026, may alter the qualification architecture, assessment rules and implementation timetable for senior-secondary subjects. Applied Intelligent Systems should therefore be regarded as an approved direction under development, not a settled qualification pathway. (ncea.education.govt.nz)

Second, schools and kura are expected to continue implementing refreshed curriculum content, with all Years 0–10 content scheduled to be available by 9 September 2026 and Years 9–10 implementation beginning from 2027. This creates an opportunity to embed AI literacy into learning areas rather than treating it only as a specialist subject. (education.govt.nz)

Third, tertiary providers are likely to expand institutionally managed AI tools, particularly agents integrated with learning-management systems and student services. The emerging question will be whether these deployments are evaluated consistently enough to distinguish useful educational systems from attractive but low-value automation.

For stakeholders, the most useful indicators to monitor are:

• The proportion of schools with practical, current AI policies.
• Participation in approved tools and professional learning.
• Student and whānau understanding of acceptable use.
• Privacy and cultural-assurance processes completed before deployment.
• Changes in teacher workload after verification and oversight are included.
• Assessment outcomes and authenticity incidents.
• Independent evidence of effects on learning and equity.
• The final treatment of AI within the replacement qualification system.

Overall Assessment

AI in New Zealand education is widely adopted in practice but not yet systematically governed.

The previous edition’s description of a transition from experimentation towards structured, curriculum-linked implementation remains accurate. The new ERO evidence adds a necessary qualification: structure is developing unevenly, and most schools are still relying on local interpretation, individual teacher judgement and informal student practice.

The strongest current deployments share four characteristics:

• A clearly bounded use case.
• Integration into an existing educational workflow.
• Human responsibility for consequential decisions.
• Deliberate attention to privacy, assessment integrity and cultural context.

NZQA demonstrates how AI can operate at national scale under controlled conditions. Unitec demonstrates that tertiary providers are beginning to build reusable institutional AI capability. Education Perfect shows how centrally funded platforms may introduce AI into everyday schooling. Applied Intelligent Systems signals that AI capability is moving into formal curriculum planning, although it remains a future subject rather than a current deployment.

The principal risk is not that education will fail to adopt AI. It is that adoption will become widespread without equally widespread capability to judge when AI is educationally appropriate.

New Zealand’s immediate priority is therefore coherence rather than acceleration: a national framework, clearer guidance, sustained professional learning, culturally legitimate data practices, equitable access and assessment systems that continue to show what learners themselves know and can do.