AI in Education in New Zealand: A Living Whitepaper

Updated for publication on 13 July 2026

Executive Snapshot

AI in New Zealand education is now best understood as embedded, governed more explicitly, and still unevenly operationalised. Since the previous update on 10 June 2026, the most important movement has not been a dramatic jump in classroom use; it has been the thickening of system infrastructure around AI: new Budget 2026 funding for NZQA pilots, further senior curriculum consultation activity, stronger institutional assessment redesign in universities, and more visible signposting of AI rules inside the NCEA ecosystem. (education.govt.nz)

The adoption picture remains broad. TALIS 2024 still provides the strongest national benchmark: 69% of New Zealand Year 7–10 teachers had used AI in the previous year, versus 36% across the OECD lower-secondary average, while the most common uses remained planning and summarisation rather than marking. At the same time, teachers continue to report strong demand for professional learning in AI. (educationcounts.govt.nz)

The sector’s current centre of gravity is increasingly clear. In schools, official guidance still frames AI as teacher-supporting, not teacher-replacing, especially in assessment. In tertiary education, leading institutions are redesigning assessment architectures, formalising AI tool access, and building governance models that explicitly include privacy, Te Tiriti, and data sovereignty. At system level, NZQA remains the most mature production user, and is now backed by fresh public funding to expand AI and machine-learning pilots in assessment design, delivery, and moderation. (education.govt.nz)

What Has Changed Since 10 June 2026

• Budget 2026 added direct funding for AI pilots in national assessment. The Ministry says Budget 2026 includes $2.1 million operating funding for NZQA to develop pilots and proofs of concept for automated text scoring, AI, and machine learning to support marking, moderation, and exam development in NCEA and New Zealand Scholarship. (education.govt.nz)
• NZQA has now baked AI exploration into its 2026/27 forward plan. In its new Statement of Performance Expectations for 1 July 2026 to 30 June 2027, NZQA says it will continue to grow digital assessment offerings and explore emerging technologies such as AI to support assessment design and delivery. (www2.nzqa.govt.nz)
• Senior secondary curriculum consultation moved into Group 3. Tāhūrangi’s Phase 5 consultation page, updated 9 July 2026, says Group 3 subjects are open for feedback from 6 July to 3 August 2026, while four further subjects still require more development before release. (newzealandcurriculum.tahurangi.education.govt.nz)
• The NCEA site now has a dedicated GenAI page. As of 12 July 2026, NCEA.education includes a page titled “Generative AI in NCEA assessment”, but it currently does little more than point users back to Ministry guidance. That makes it a visibility and navigation change rather than a substantive policy change. (ncea.education.govt.nz)
• University AI strategies have become more publicly legible. Since late May and June, AUT has published a clearer public-facing AI strategy and assessment model, while the University of Canterbury has pushed AI literacy further into formal framework-building, short courses, student support, and postgraduate programme design. (aut.ac.nz)

Overall State of Adoption

Schools and kura: mainstream use, patchy readiness

The best national picture remains TALIS 2024. It shows AI use is already mainstream among many teachers, but mostly for lesson planning, content support, and summarisation, not high-stakes assessment judgment. Among New Zealand teachers who had used AI, 78% had used it to generate lesson plans and activities and 73% to learn about or summarise a topic, while only 12% had used it to assess or mark student work. Professional-learning demand remains high, especially in secondary settings. (educationcounts.govt.nz)

NZCER’s primary-school AI research still sharpens that picture. Its 2025 report, based on end-2024 surveys, found widespread experimentation by teachers and substantial student exposure, but also significant gaps in policy clarity, premium tool access, and confidence. Three-quarters of responding teachers had no school-funded premium AI access, fewer than half felt confident teaching responsible AI use, and 85% wanted more training. Students commonly used AI more outside school than inside it and were often unsure what their school rules were. (nzcer.org.nz)

The Ministry’s current guidance remains explicitly augmentation-first. It says AI should improve teaching and learning by supporting knowledgeable and skilled teachers, that human oversight is essential, and that AI used for marking must support rather than replace teacher professional judgment. It also reiterates that schools with consent to assess must have authenticity policies that include acceptable GenAI use. (education.govt.nz)

Assessment is the main governance battleground

The strongest policy movement is still around assessment. Ministry guidance says AI-assisted marking is more suited to assessment for learning than to complex or summative judgment, and the companion marking guidance says AI use to mark NCEA internal assessments is discouraged and should be limited to supporting the assessor’s final judgment. (web-assets.education.govt.nz)

This is increasingly consistent with what universities are doing. The University of Auckland’s Two-Lane Approach requires all courses and programmes to implement the model by 2027, distinguishing between controlled assessments where AI is restricted by default and uncontrolled assessments where AI use may be unrestricted. AUT has adopted a similar institution-wide two-channel approach, with Channel 1 for invigilated, non-AI summative judgments and Channel 2 for AI-permitted settings, supported by its “Points of Observation of Learning” model. (teachwell.auckland.ac.nz)

The implication is now hard to miss: in New Zealand, the credible response to AI is shifting away from detector-led enforcement and toward assessment redesign, disclosure, observation, and staged assurance of learning. That is visible across Ministry, AUT, and Auckland materials alike. (web-assets.education.govt.nz)

Tertiary education: operating models are maturing

The University of Auckland remains one of the clearest examples of AI codification in teaching and assessment. Its Two-Lane model is now an explicit institutional position, and its public AI-in-education hub continues to push course-level implementation and second-half 2026 Cogniti deployments. Cogniti is integrated into Canvas and positioned as a bounded, course-specific AI environment rather than a free-for-all external chatbot. (teachwell.auckland.ac.nz)

AUT has become more publicly explicit as well. Its published AI principles emphasise equity, integrity, Māori and Indigenous data considerations, human-centred deployment, and the irreplaceable value of human-to-human connection in education. In learning and assessment, AUT says it is taking a whole-of-institution approach to redesign, implementing new assessment principles through 2025 and 2026 and openly treating AI adaptation as a core strategic project rather than a peripheral teaching issue. (aut.ac.nz)

The University of Otago remains the clearest example of institution-level governance maturity. Its AI Governance Policy, effective 10 March 2026, applies across development, procurement, deployment, and use of AI systems; explicitly references Te Tiriti obligations, Māori data sovereignty, Pacific values, and environmental impact; and requires risk assessment and approval before AI is used on confidential or restricted data. (otago.ac.nz)

The University of Canterbury is also broadening the tertiary picture. UC’s June 2026 SAIL announcement frames AI literacy as more than tool use, stressing critical judgment, citizenship, and inclusion of Māori and Pacific voices. UC has also embedded AI in student support and governance through endorsed tools, AI skills workshops, and formal postgraduate offerings, including a new Master of Applied Artificial Intelligence with an education pathway and a new Master of AI and Education referenced in UC’s own reporting. (canterbury.ac.nz)

System agencies: NZQA is still the most mature production user

NZQA’s use of AI in co-requisite literacy assessment remains the most consequential live deployment in the sector. NZQA’s current marking page says that from May 2025, it began using a combination of automated machine marking, automated text scoring, and human marking for literacy and numeracy co-requisites, with automated text scoring applied to all digitally submitted Writing assessments and human check-marking used for boundary cases. (www2.nzqa.govt.nz)

NZQA’s public material continues to describe the May 2025 deployment as a major milestone: more than 55,000 writing assessments were marked with automated text scoring, results came back 3.5 weeks earlier, and human check-marking remained in place for over a third of responses near the achievement boundary. NZQA also says its AI work is guided by public-service principles including transparency, safety, accountability, and human-centred values. (www2.nzqa.govt.nz)

What is new is not only that NZQA is already using AI at scale, but that the Government is now funding the next layer: pilots for AI and machine learning in marking, moderation, and exam development, alongside new NZQA planning language about using AI in future assessment design and delivery. (education.govt.nz)

Policy, Governance, and Trust

Official school guidance is stable, but more operational than before

There has been no major new Ministry AI policy shift since the 22 May 2026 guidance refresh, but that guidance remains one of the most operationally important documents in the sector. It gives schools concrete boundaries on marking, personal-information handling, age restrictions, school-policy design, and authenticity expectations for assessment. (education.govt.nz)

A small but telling July development is that AI is now easier to find inside the NCEA support environment itself. The dedicated NCEA GenAI page does not yet add new policy, but it signals that AI in assessment has become a standing issue in national qualifications support rather than a side topic. (ncea.education.govt.nz)

Privacy, children’s rights, and Māori data concerns are intensifying constraints

The privacy environment is becoming more important, not less. The Privacy Commissioner’s 2026 annual survey, published 11 May 2026, found that 71% of New Zealanders were concerned about children’s privacy and 67% were concerned about agencies or businesses using AI to make decisions about people using personal data. Māori respondents showed higher concern and lower trust on many measures. (privacy.org.nz)

The Commissioner’s new education-sector children’s privacy guidance makes the implications for schools more concrete. Chapter 16 says AI use in education must comply with the Privacy Act, warns about risks such as vendors using learner data to train models, profiling, inappropriate chatbot interactions, and overcollection of personal information, and recommends careful due diligence before new tools are introduced. Separate AI guidance also says organisations should conduct privacy impact assessments and engage with Māori about risks to taonga information. (privacy.org.nz)

This means that in New Zealand, trust, privacy, and cultural legitimacy are no longer secondary implementation issues. They are now core constraints on how AI can scale in education. (privacy.org.nz)

Research Overview

National evidence: teachers are ahead on use, not yet on capability coherence

TALIS 2024 remains the strongest official evidence base for system-wide teacher use. New Zealand teachers are ahead of many international peers in AI exposure and use, but not yet in settled capability, especially around assessment and professional learning. That combination still defines the current phase: broad familiarity, incomplete operational maturity. (educationcounts.govt.nz)

NZCER: adoption is real, but access and guidance remain uneven

NZCER’s primary-school study is still the most useful New Zealand research on how AI is actually entering classrooms. Its value lies less in national representativeness than in showing how even relatively engaged teachers experience AI: high experimentation, limited funded access, uncertain rules, and a strong call for more training and clearer frameworks. (nzcer.org.nz)

UC’s SAIL framework: AI literacy is becoming a substantive field of work

The University of Canterbury’s June 2026 reporting on the Scaffolded AI Literacy (SAIL) framework is a meaningful addition to the evidence base because it pushes AI literacy beyond “how to prompt” and toward staged understanding, critical judgment, and AI citizenship across age groups and contexts. UC also says the work has informed short courses and its Master of AI and Education, while Professor Kathryn MacCallum is engaging with both the Ministry and ERO. (canterbury.ac.nz)

ERO is still studying the issue, not yet closing it

ERO’s evidence programme still lists “How are schools and students using AI?” as an active question. As of 13 July 2026, that means national evaluation is still catching up with practice rather than fully defining it. (evidence.ero.govt.nz)

Case Studies

1) NZQA: AI at national assessment scale

NZQA remains the strongest case study of operational AI in education because the deployment is real, national, and quality-assured. It combines automated scoring with human check-marking, is already embedded in live literacy assessment operations, and is now being extended through government-backed pilots into adjacent areas such as moderation and exam development. (www2.nzqa.govt.nz)

2) University of Auckland: bounded course AI plus assessment redesign

The University of Auckland offers the clearest combination of assessment model plus tooling. The Two-Lane Approach gives a whole-of-programme logic for when AI should be restricted or permitted, while Cogniti offers a controlled course-based AI environment inside Canvas. That pairing is important because it moves beyond abstract principles to practical delivery architecture. (teachwell.auckland.ac.nz)

3) AUT: whole-of-institution assessment redesign

AUT’s case is notable because it is not treating AI as a local teaching innovation but as a university-wide redesign challenge. Its two-channel model, AI principles, and phased implementation of new assessment policy suggest a more systemic response focused on assurance of learning rather than prohibition. (aut.ac.nz)

4) University of Otago: governance-first institutional maturity

Otago’s AI Governance Policy is a case study in whole-organisation readiness. It builds AI into formal committee structures, risk management, data protection, Te Tiriti obligations, and staff capability expectations. In the New Zealand context, that is a significant marker of institutional maturity. (otago.ac.nz)

5) Ministry case studies: school-level authenticity redesign

The Ministry’s Aotea College and Hobsonville Point Secondary School case studies remain important because they show how schools are responding without relying on AI detectors. The approaches include checkpoints, common templates, authenticity criteria, verbal checks, and policy changes designed to make student process more visible. (education.govt.nz)

Emerging Trends

1) Augmentation remains dominant

Across schools, tertiary institutions, and NZQA, the dominant New Zealand model is still AI as support: planning, tutoring, feedback, workflow assistance, marking support, and service navigation, with human judgment retained at the final decision points. (education.govt.nz)

2) Assessment redesign is hardening into doctrine

The sector is converging on a shared answer to AI: redesign assessment rather than pretend AI can be excluded from all learning contexts. Ministry guidance, Auckland’s Two-Lane model, and AUT’s two-channel approach all reinforce that shift. (web-assets.education.govt.nz)

3) AI literacy is scaling faster than formal curriculum change

The quickest movement is in AI literacy frameworks, resources, workshops, and local operating models, not in a standalone compulsory AI subject. That is visible in Ministry resources, UC’s SAIL framework, AI Forum’s Blueprint emphasis on talent and literacy, and ongoing curriculum consultation that embeds AI in subjects such as Digital Technologies and Computer Science. (newzealandcurriculum.tahurangi.education.govt.nz)

4) The next battleground is implementation quality

The key divide is no longer between adopters and non-adopters. It is between institutions with coherent governance, funded tools, staff development, and assessment redesign capability, and those still relying on ad hoc teacher improvisation. TALIS, NZCER, and the contrasting maturity of university models all point in that direction. (educationcounts.govt.nz)

5) New Zealand’s distinctive feature remains cultural and privacy legitimacy

Te Tiriti obligations, Māori data sovereignty, children’s privacy, and community trust are not side issues in New Zealand’s AI-in-education story. They are part of the design brief itself, and they are becoming more prominent across policy, university governance, and public-attitudes data. (otago.ac.nz)

Conclusion

As of 13 July 2026, AI in education in New Zealand is best characterised as normalised in use, increasingly structured in governance, and still uneven in execution. The most meaningful recent developments since 10 June 2026 are system-level rather than sensational: Budget 2026 funding for NZQA AI pilots, forward planning inside NZQA, continued Phase 5 curriculum consultation, and more explicit university-wide models for assessment and AI governance. (education.govt.nz)

The overall trajectory is clearer now than it was a month ago. New Zealand is not moving toward an “AI replaces educators” model. It is moving toward a model in which AI supports teaching, literacy, service delivery, and selected assessment operations, while human judgment, authenticity, privacy, and cultural legitimacy become more explicit conditions of use. (education.govt.nz)

The defining challenge from here is implementation quality at scale. New Zealand has already crossed from experimentation into operational adoption, but a fully coherent national operating model still does not exist. The organisations furthest ahead are those treating AI not as a gadget question but as a whole-system education design problem spanning pedagogy, assessment, governance, data, equity, and trust. (educationcounts.govt.nz)