Ministry of Education AI Guidelines

This is my second critique of a government agency’s AI policy, following my case study of the Ministry for Regulation’s 2026 AI guidance for regulators.

The Ministry of Education’s internal guidelines for staff use of Artificial Intelligence and generative AI were released under the Official Information Act 1982 in September 2025. The four-page document, titled Ministry Guidelines for using Artificial Intelligence and Generative AI, tells staff which AI tools they may use, how to classify information before entering it, and which risks to watch for.

Te Tiriti o Waitangi does not appear anywhere in the document. Neither do the words Māori, iwi, hapū, whānau, te reo Māori, mātauranga Māori or tikanga. Māori data and Māori data sovereignty are also not mentioned.

This is the Ministry responsible for the education of tamariki Māori, administering an Act whose stated purpose includes an education system that honours Te Tiriti o Waitangi. Below I set out what the guidelines say, where they fall short, and how they measure against the Māori AI frameworks I have published since 2020.

What the guidelines say

In summary:

  • Microsoft Copilot Chat is the only generative AI tool approved by the Ministry. It may be used for information marked UNCLASSIFIED or IN-CONFIDENCE, or that is in the public domain.
  • ChatGPT, Grok, Gemini, Perplexity and Claude are blocked on Ministry devices.
  • Staff must not enter information classified as SENSITIVE or RESTRICTED, information that has not yet been classified, or personally identifiable information such as a name, date of birth or address.
  • Security and privacy, accuracy, and bias. On bias, staff are told to “always review results for fairness.”
  • Staff must not use AI to make decisions, especially about public services, and are encouraged to have AI-generated content checked by at least one other person.
  • New tools go through a staff member’s manager, the Service Desk, and the Matihiko Technology Needs Assessment process.
  • Ethical questions go to an Ethics Advisory email address.

Keeping humans responsible for decisions, refusing to automate public services, and requiring peer review of AI output are all good practice. But the authority and training of the humans in the loop also raises many ethical questions and concerns.

In addition to that;

Te Tiriti o Waitangi

Section 4 of the Education and Training Act 2020 states that the purpose of the Act includes establishing an education system that “honours Te Tiriti o Waitangi and supports Māori-Crown relationships.” Section 14 of the Public Service Act 2020 states that the role of the public service includes supporting the Crown in its relationships with Māori under Te Tiriti.

Staff guidance on AI is where those obligations meet daily practice. It decides what data goes into which systems, who approves new tools, and whose interests are checked before that happens.

The Human Rights Commission 2026 report AI and Digital Technologies: A Human Rights and Te Tiriti o Waitangi Approach, said the government has primary responsibility for giving effect to Te Tiriti through law, policy, investment and public-service delivery.

Measured against the 6 Te Tiriti Based AI Ethical Principles

In 2023 I published the 6 Te Tiriti Based Artificial Intelligence Ethical Principles for any AI system that uses, produces or makes decisions about Māori data. They are intended as a starting point for exactly this kind of document: an organisation’s AI policy. The table below applies each principle to the Ministry’s guidelines.

Principle What it asks of an AI policy Ministry of Education guidelines
Tino Rangatiratanga Māori authority over Māori data and over AI that affects Māori. Authority sits with managers, the Service Desk and the Matihiko assessment process. Whānau, hapū, iwi and Māori communities have no role.
Equity AI must not widen existing gaps in outcomes for Māori. One line: “review results for fairness.” No recognition that bias in these tools falls most heavily on Māori learners and te reo Māori.
Active protection The Crown actively protects Māori interests in data. Protection is defined only by security classification. Māori data receives no protection beyond what any other data receives.
Mana whakahaere Māori take part in the governance of AI and the decisions made about it. No Māori role in tool approval, ethics advice or review of AI output.
Mana Motuhake Māori are able to build, own and control our own AI. One approved tool, a general offshore product. No pathway that recognises Māori-built tools or Māori-governed data.
Tapu/Noa Culturally safe practice, including recognising knowledge that should not go into AI systems. Information is graded by its sensitivity to the Ministry, not by tikanga. Whakapapa, karakia or mātauranga marked “public domain” could be entered into Copilot without any check.

 

The Ministry’s guidelines fail all six principles because they do not engage with them at all.

I also wrote an AI and Generative AI Policy template for education which is much more protective than the official guidelines.

I use an architecture-versus-decoration test when reviewing policies like this one. It asks whether Māori provisions are structurally operative, shaping who decides and what happens, or decorative, present in the text but changing nothing. Many government AI documents fail that test because their Te Tiriti statements are decoration. These guidelines do not get that far. There is no decoration to test.

Security classification is not data sovereignty

The guidelines classify information by how sensitive it is to the Ministry, then decide which tool it can go into. That framework has no place for Māori data, because Māori data is not defined by its security marking.

Information can be UNCLASSIFIED and still be Māori data. Achievement data for a kura, attendance patterns in a rohe, te reo Māori teaching resources and reports on Māori-medium education may all be cleared for Copilot under these rules. None of them would be checked for Māori interests first.

In my 2020 Treaty of Waitangi/Te Tiriti and Māori Ethics Guidelines for AI, Algorithms, Data and IoT, written mainly for government and its contractors, I warned of the heightened risk of reidentification when anonymised data comes from small populations. The Ministry’s rule against entering names, dates of birth and addresses protects individuals. It does not protect collectives. Data about a small kura or a single hapū community can identify whānau without containing a single name.

The Waitangi Tribunal’s WAI 2522 report recognised Crown obligations to actively protect Māori interests in data. Logging in to Copilot with a Ministry on premise account is a security control. It answers the question of whether data is safe from outsiders. It does not answer who holds authority over Māori data once it is inside a Microsoft system. Data residency and security are not the same as data sovereignty.

He Tangata, He Karetao, He Ātārangi

My kaupapa Māori AI framework, He Tangata, He Karetao, He Ātārangi, describes AI as a person, a puppet and a shadow. It explains why accountability for AI is easily diffused, and why the system itself can never be the party held to account.

Read through that framework, the guidelines show both a strength and a weakness.

The strength is that the Ministry keep decisions with people. Staff must not use AI to make decisions about public services, and AI output must be checked. That is consistent with the framework’s position that accountability must stay with people, not systems.

The weakness is the guidelines tell each staff member, “You are responsible for keeping information secure.” Responsibility is placed on the individual public servant. It is not placed on the Ministry as a Treaty partner.

Copilot is he karetao, a puppet animated by forces outside the Ministry: Microsoft’s design choices, training data and commercial priorities. It is also he ātārangi, a shadow that carries the mauri of whatever is fed into it, including Māori data. A guideline that asks individual staff to manage those risks alone, with no Māori data governance behind them, spreads accountability thin in exactly the way the framework warns against.

Part of a wider pattern across government

In 2025 I sent Official Information Act requests to 31 government departments and a number of Crown agents and other agencies about their Māori data and data sovereignty policies. Of the 24 agencies that had a data strategy or policy, eight did not recognise Māori or Te Tiriti rights at all. Also, the governments National AI Strategy itself makes no mention of Te Tiriti.

The contrast within government is telling. Te Puni Kōkiri‘s AI policy, released under the OIA in May 2025, lists Tiriti-based AI principles of tino rangatiratanga, equity, active protection, mana whakahaere, mana motuhake and tapu/noa. These are the six principles I published in 2023.

I maintain an evolving public list of 70 unique Māori AI and data governance frameworks developed by Māori researchers, iwi, hapū, Māori organisations and the government itself. The Ministry did not need to invent anything. It needed to use what is already there.

What the Ministry should do

The guidelines say they “will continue to adapt alongside these advancements and public expectations.” The next revision should:

  1. Start from Te Tiriti and state the Ministry’s obligations under section 4 of the Education and Training Act 2020 and section 14 of the Public Service Act 2020, and treat Te Tiriti as a constitutional threshold, not a consultation step to tick off.
  2. Add Māori data as a category alongside security classifications, so that a Māori data check is required no matter how the information is marked.
  3. Include Māori data governance criteria in the Matihiko Technology Needs Assessment and in ethics advice, with Māori expertise involved in both.
  4. Set out which kinds of knowledge, such as whakapapa, karakia and mātauranga held by iwi and hapū, must not be entered into AI tools regardless of classification.
  5. Replace the single fairness line with specific guidance on bias against Māori learners, whānau and te reo Māori, including checks on how AI tools handle te reo Māori.
  6. Where Māori data relates to particular iwi, hapū or rohe, decisions about its use in AI should involve those groups, not only internal teams.
  7. Keep human decision-making, but place responsibility for Treaty compliance on the Ministry, not on individual staff members.

Tamariki Māori will grow up with AI built into their schooling. The Ministry that sets the rules for that system should be able to show how its own use of AI honours Te Tiriti. At the moment Māori are invisible.

How to cite this page

Taiuru, K.N. (2026, November 18). Ministry of Education AI Guidelines. Taiuru & Associates Ltd. https://www.taiuru.co.nz/ministry-of-education-ai-guidelines/