Ticking the Box: What an OIA Release Shows About Māori Expertise in Government AI Guidance
A case study on the Ministry for Regulation’s 2026 guidance for regulators using artificial intelligence
Sumamry
- In May 2026, the Ministry for Regulation (Te Manatū Waeture) published guidance for regulators on using artificial intelligence (AI).
- Before publishing, the Ministry paid an AI consultancy for an “independent expert peer review”. No Māori reviewer, and no Māori data governance expert, was involved.
- The AI consultancy itself warned that the Māori content risked becoming “lip-service and sign posting”, and strongly recommended that the Ministry engage Māori experts.
- The Ministry decided not to, describing that work as “outside our skillset” and as “adding delay”.
- The lesson is simple: Māori expertise must be part of the AI policy life cycle from the start, not an amendment at the end.
Key learnings from this case study
Across New Zealand, two patterns are well documented. Employers lean on their Māori staff to provide Māori representation, and government agencies rely on Māori groups that do not hold the authority or the relevant skills. Neither problem is new, but both are growing as agencies introduce AI systems.
AI Forum statistics show an increase in organisations seeking Māori expertise and input into AI projects. What is less clear is whether that is genuine engagement.
Internationally, there are many examples of AI being developed without involving the communities it affects. The consequences include bias, exclusion, harm to communities, human rights impacts, system failures and damage to an organisation’s reputation.
This case study shows how the same pattern can play out inside our own public service, and what doing it properly would look like. It is written to illustrate a recurring failure pattern in public-sector AI policy, not to single out individuals or organisations. Names appear only because they appear in the documents. Readers are encouraged to weigh the evidence, debate it and draw their own conclusions.
Where this information comes from
Unless stated otherwise, everything in this case study comes from documents released by the Ministry for Regulation under the Official Information Act 1982:
- OIA request
- LensenMcGavin AI peer-review engagement and report
- Internal feedback and decision emails
The case study relates to Version 1 of Practical steps for regulators: leading AI innovation with confidence. A second version has since been published (see What happened next).
Timeline
| Date | Event |
|---|---|
| 5 February 2026 | Internal feedback email sets out why an external review is wanted and how Māori data sovereignty will be handled. |
| Early 2026 | The Ministry engages LensenMcGavin AI to peer-review the draft. Its report recommends engaging Māori experts. |
| 12 April 2026 | The lead official sets out two options for the Māori content and prefers keeping it high-level. |
| 14 April 2026 | The outcome is settled: a Te Tiriti section without depth on Māori data sovereignty, to be tested later with Te Puni Kōkiri only. |
| May 2026 | Version 1 of the guidance is published. |
| After publication | Public critique, a meeting with the Ministry, and a revised Version 2. |
What the documents show
1. The review was wanted so the Ministry could say a review had happened
The earliest document is an internal email from one official to a colleague. It is candid about the purpose of an external review:
“…it helps to say this has been external reviewed by XX. This is good practice which is why we have done something similar with each of our sector review reports…”
— Appendix C, internal email, 5 February 2026
In plain terms, one purpose of the review was to be able to say a review had taken place.
2. Māori data sovereignty was put in the too-hard basket
The same email sets out how the Ministry intended to handle the Māori content:
“I do not believe Māori data sovereignty is well understood across the public service yet and we do not want to try and define it in the document. However, I suggest we add some links to the authoritative source… as the crown have defined it.”
— Appendix C, internal email, 5 February 2026
There are three problems here.
- Low understanding was treated as a reason to avoid the subject, rather than a reason to bring in people who do understand it.
- The “authoritative source” was to be the Crown. The Crown is the Treaty partner whose exercise of power Māori data sovereignty exists to constrain. Asking the Crown to define it turns the concept upside down. Tino rangatiratanga over Māori data does not come from a Crown definition of it.
- The Crown has no single, agreed definition. Meanwhile, the Waitangi Tribunal has already given clear direction in WAI 2522, recognising Māori data as a taonga.
3. The “independent expert” had no Māori expertise
The Ministry engaged LensenMcGavin AI, which describes itself as an AI firm (Appendix B). Its critique of the draft was that some passages:
“come across as very ‘ChatGPT'”.
The consultancy’s own report includes this disclosure:
“Generative AI tools (e.g. Gemini) were used to refine writing and structure. … All suggestions, critiques, and comments were produced by the authors. All GenAI output was reviewed and edited by the authors.”
A later courtesy review of the revised draft reads:
“No notes from our end. Absolutely brilliant… without wanting to sound too much like a sycophantic chatbot myself… you will have produced the gold standard for public sector AI guidance.”
Whatever view readers take of the reviewer’s independence, the more important point is this: the engagement did not provide the Māori-specific expertise that was later identified as necessary.
The contract scoped the work as feedback:
“Suitable for a senior regulatory audience with a non-technical or IT background or capability”
— Appendix B
By its own terms, this was a light-touch review of readability and framing. It was never a specialist review of the document’s most consequential parts.
4. The AI reviewer itself said to bring in Māori experts
The clearest warning about the Māori content came from the AI reviewer. Section 4.6 of its report states:
“There is a real risk that in their current state, the concepts will function merely as lip-service and sign posting… Māori will disproportionately feel the negative impacts of AI… we strongly recommend that MfR engages with experts and leaders in this domain to properly inform these components of their guidance.”
— Appendix B, LensenMcGavin AI peer review, s 4.6
That this came from a non-Māori AI consultancy is itself the point. The Ministry was willing to pay for, and act on, an external voice. It chose the wrong external voice for this content. When that voice pointed to the right one, the Ministry did not follow it.
5. The recommendation was treated as a cost, not an obligation
On 12 April 2026, the lead official framed the choice like this:
“1. Keep this guidance high-level and refer to system guidance (preferred and as currently drafted), or 2. Expand this section further… recognising that further expansion is outside our skillset and would require consultation by GCDO and TPK adding delay to publication of this guidance.”
— Appendix C, internal email, 12 April 2026
Proper engagement was described as “outside our skillset” and as “adding delay”.
But “outside our skillset” is exactly why external Māori expertise was needed. It is the same reasoning the Ministry used to buy in AI expertise, because it lacked technical capability. The Ministry applied buy-in logic to the technical gap and avoidance logic to the Te Tiriti gap.
On 14 April, the outcome was settled. The guidance would include a Te Tiriti section, but would not go deep into Māori data sovereignty “at this stage”. The draft would be tested with Te Puni Kōkiri only. That is consultation with a Crown agency on the Crown’s own drafting, after the fact. It is not Māori partnership in shaping the guidance.
Why this approach falls short
Te Tiriti is a threshold, not a refinement
The Ministry treated the Māori content as one improvement among many, weighed against timeline, skillset and scope, much like a grammar fix or a missing case study.
Te Tiriti o Waitangi is not an optional enhancement. It is a constitutional relationship that conditions how the Crown may design and deploy decision-making systems. It must be met first, not traded off.
Architecture or decoration?
A simple test separates genuine engagement from the appearance of it:
- Architecture: Are the Māori provisions structurally operative? Do they change what a regulator must actually do?
- Decoration: Or are they there so the document can be seen to mention them?
The AI reviewer, without using these words, reached the same conclusion when it warned of “lip-service and sign posting”. The Ministry’s chosen path (keep it high-level, link to Crown sources, avoid a definition, defer the depth) is the path that produces decoration.
Decoration can only be fixed by making Māori expertise part of the life cycle and the review.
Proper engagement places authority with those who hold it: hapū, hāprori Māori, iwi and Māori data governance experts. They need to be involved at the very start, while the scope is still being set, not brought in so late that all they can do is react.
Data residency is not data sovereignty
The guidance and its review both circle around where data is stored. These are two different questions:
- Data residency is about servers and jurisdiction: where the data physically sits.
- Māori data sovereignty is about authority: who decides how Māori data is collected, used and governed, and on whose terms.
An AI reviewer scoped for a non-technical audience was not positioned to hold that distinction, and the record shows it did not. Māori expertise reliably does. Procuring the wrong reviewer produced the wrong frame.
Those most at risk were engaged least
The review itself stated that “Māori will disproportionately feel the negative impacts of AI.”
Yet the process produced an inversion. The group facing the most harm received the least engagement, because engaging them would take longer. Despite the reviewer’s warning, publication speed appears to have been given priority.
Kaupapa Māori researchers and Māori leaders regularly raise concern that “Māori guidance” is becoming disconnected from tikanga and Māori values, sometimes even when Māori groups are involved. Version 1 of this guidance went further still. In its development and review, no iwi, hapū, rōpū, Māori organisation, Māori data governance body or individual Māori expert was consulted at all.
What proper engagement could have looked like
| What the Ministry did | What proper engagement requires |
|---|---|
| Commissioned a general AI consultancy to review the whole document, including the Māori content. | Commission a separate, parallel review of the Te Tiriti and Māori data sections by a reviewer with Māori data governance expertise. |
| Sought external review so it could say a review had occurred. | Seek review to improve the document, and choose reviewers whose mana makes the content right, not merely defensible. |
| Treated “outside our skillset” as a reason to stay high-level. | Treat “outside our skillset” as the trigger to bring in people whose skillset it is, the same logic used to justify buying AI expertise. |
| Deferred depth on Māori data sovereignty to avoid delaying publication. | Plan the work so Māori engagement shapes the scope from the start. Publication timing follows the obligation, not the other way round. |
| Planned to test the finished drafting with Te Puni Kōkiri, a Crown agency. | Engage Māori data governance expertise at the start of policy development, with Te Puni Kōkiri and other Crown Māori capability providing institutional support where relevant. |
| Let the draft use words like “taonga” and “sovereignty” as vocabulary. | Let Māori expertise make those words structurally operative, changing what a regulator must do, not just how the guidance reads. |
A simple Māori AI governance test
Any agency can ask these five questions of its own AI guidance or project:
- Is Māori content merely present?
- Has Māori expertise informed it?
- Can Māori expertise change the substance?
- Does the guidance require regulators to do anything differently?
- Who decides whether those requirements have been met?
If the honest answer to the first question is “yes” and the rest are “no” or “unclear”, the Māori content is decoration.
For a fuller framework, see my He Tangata, He Karetao, He Ātārangi framework: https://www.taiuru.co.nz/kaupapa-maori-ai-framework/
Lessons for AI more broadly
AI systems can reproduce and amplify existing inequalities. AI policy and guidance have often failed to properly include Māori principles, even though Māori are likely to feel the negative impacts of AI most. The risks show up at several levels.
- Data: Māori data, and data carrying bias, can end up in datasets without proper regard for Māori authority, where the data came from, or how it is governed.
- Models: AI models can encode historical patterns and biases that harm Māori.
- Procurement: Agencies can buy AI systems without enough visibility of the training data, testing data, model development or vendor practices.
- Decision-making: A human-in-the-loop can give the appearance of human control without the decision-maker having the technical, cultural or subject expertise to exercise it. The reviewer explicitly warned that human-in-the-loop is not enough without review and audit processes that address human error and automation bias. Having a person in the loop is not the same as having someone accountable for the loop.
- Accountability: The guidance itself says accountability stays with organisations and legally authorised decision-makers. The reviewer thought this point important enough that it should be made much more prominent.
A human decision-maker cannot meaningfully oversee an AI system if they do not understand its limitations, where its data came from, the risks in its outputs, and the cultural and legal context in which those outputs are used.
Lessons for government AI procurement
The Ministry’s contract was carefully scoped around the needs of senior regulatory leaders: regulatory defensibility, clarity, practical use and technical accuracy. But the subject matter needed more than that.
Government AI procurement should ask one question at the very start:
What expertise do we need to properly evaluate this AI system or policy?
The answer may include:
- AI technical expertise
- information security
- privacy
- legal expertise
- subject-matter expertise
- Māori data governance expertise
- Te Tiriti expertise
- accessibility expertise
- knowledge from affected communities.
The right expertise should be set by the subject matter and the level of risk, not by the procurement category. Otherwise an agency can technically meet a requirement for “independent peer review” while still missing the specialist scrutiny its most important content needed.
Conclusion
The Ministry for Regulation’s AI guidance is an unusually clear example of the difference between acknowledging Māori interests and embedding Māori expertise.
The Ministry commissioned an independent review of its AI guidance. That review raised many technical, governance and regulatory issues. Among them was a specific and substantive concern about Māori data, Māori data sovereignty and Te Tiriti o Waitangi.
The reviewer did not simply suggest adding a paragraph. It warned that the existing approach risked becoming “lip-service and sign posting”, and strongly recommended engaging Māori expertise.
The Ministry went on to include Te Tiriti considerations in the near-final guidance, but said the material would be expanded later, following sector feedback and support from Te Puni Kōkiri.
Which leaves one question. If Māori data sovereignty and Te Tiriti were important enough to appear in government AI guidance, and an independent reviewer specifically said Māori experts were needed, why was that expertise not built into the development and review process from the outset?
Māori expertise should not be an amendment to AI governance. It should be part of its life cycle.
What happened next
This section draws on my own experience after publication, not on the OIA documents.
After Version 1 was published, I criticised the guidance and the independent review on LinkedIn. Reactions were mixed, though largely supportive of the critique.
The Ministry contacted me the next day and we met to discuss the issues. The Ministry was open to feedback, and the written feedback I provided was included in Version 2, which is the current version at the time of writing.
Even so, this case study and its critique remain relevant. The guidance still lacks proper Māori perspectives and a full acknowledgement of Te Tiriti.
Both the Ministry and the consultancy were given this case study in draft months before publication. The Ministry thanked me for it. The consultancy was not happy about its publication.
How to cite this page
Taiuru, K. (2026, November 1). Ministry for Regulation 2026. Taiuru & Associates Ltd. https://www.taiuru.co.nz/need-for-maori-in-ai-governance-and-policy/