NZ government AI: moving from strategy to operational delivery
New Zealand now has a national AI strategy, a Public Service AI Framework, responsible-use guidance and a growing pipeline of public-sector use cases. The next challenge is converting that foundation into governed services, workflows and decisions that create measurable public value.
The current NZ government AI landscape
New Zealand has moved beyond the point where public-sector artificial intelligence is only an emerging policy topic. Agencies now have national strategy, public-service guidance, a cross-agency framework and an expanding body of practical use cases.
The Government Chief Digital Officer’s 2025 survey recorded 272 AI use cases across 70 government organisations. Of those, 55 were reported as deployed and in operational use, compared with 15 operational use cases in 2024.
That is meaningful progress. It also shows that most identified work remains outside operational delivery. The central NZ government AI challenge is therefore not whether agencies are interested in AI. It is how to move suitable use cases from ideas and trials into governed, measurable service improvement.
The central issue: frameworks establish direction, but public value is created when a governed AI use case improves a real workflow, decision or service.
Official references include the 2025 cross-agency AI survey highlights and the Government Chief Digital Officer’s wider public-service AI guidance hub.
What the 2025 NZ government AI survey shows
The survey demonstrates growing activity across public service departments and wider public-sector organisations. It also helps separate reported interest from operational adoption.
Reported activity: an idea, proposed use, experiment, trial or operational system.
Operational adoption: a system being used within an accountable process to produce a real organisational or public-service outcome.
New Zealand now has a stronger strategic foundation
New Zealand’s first national AI strategy was released in July 2025. It is adoption-focused and aims to encourage investment, reduce uncertainty and support the responsible use of proven AI technologies.
The public service also has a dedicated AI Framework and Responsible AI Guidance for GenAI. In January 2026, the Government Chief Digital Officer published a two-year Public Service AI Work Programme intended to accelerate safe and responsible uptake.
Together, these initiatives give NZ government AI work a clearer policy and governance foundation than agencies had during the earliest wave of generative AI experimentation. The remaining task is to connect that foundation to delivery capability and measurable public value.
Read the official New Zealand AI Strategy, the Public Service AI Framework and the Public Service AI Work Programme.
Why NZ government AI projects can stall before operational use
Public organisations operate under legitimate constraints. They manage personal information, statutory decisions, public money, critical services, security requirements and obligations to explain how decisions are made.
These conditions mean an agency cannot adopt AI in the same way as an individual experimenting with a consumer tool. However, risk can also create process friction that makes even low-impact use cases difficult to progress.
Good governance should enable NZ government AI adoption
Responsible governance is not the absence of AI use. It is a controlled method for deciding which uses are appropriate, what evidence is required and who remains accountable.
The official Responsible AI Guidance states that public-service GenAI use should be safe, transparent and responsible. It also emphasises human oversight because generated outputs can be misleading, harmful or biased.
For each use case, an agency should define the authorised purpose, information sources, users, system permissions, review points, records, monitoring and escalation pathway.
Define the public-service outcome
State what should improve for staff, decision-makers, service users or the public.
Assess the information and decision risk
Identify personal, confidential, classified or culturally sensitive information and the impact of an incorrect output.
Set human accountability
Name the people responsible for approving the use case, checking outputs and owning the final service or decision.
Test realistic scenarios
Evaluate normal work, incomplete information, unusual cases, poor-quality inputs and foreseeable failure conditions.
Introduce controlled operational use
Start with a bounded workflow, approved users and explicit review before expanding scope.
Measure and monitor
Track service outcomes, time, accuracy, complaints, exceptions, adoption and any unintended effects.
Privacy and accountability in NZ government AI
The Privacy Act 2020 applies when public organisations use AI to collect, store, use or disclose personal information. Existing Information Privacy Principles remain relevant even when the technology or vendor is new.
The Office of the Privacy Commissioner recommends understanding how an AI system works well enough to uphold the principles and completing a Privacy Impact Assessment before use, then updating it as the system changes.
Privacy assessment should sit alongside security, records management, procurement, legal review, accessibility, Māori data considerations and the agency’s wider public-law responsibilities.
Governance principle: the agency remains accountable for the service and its use of information, even when an external model or technology provider performs part of the processing.
See the Office of the Privacy Commissioner’s Artificial Intelligence and the Information Privacy Principles guidance.
Shadow AI is part of the public-sector adoption challenge
When approved pathways are slow or unclear, staff may experiment through personal accounts, browser tools or unassessed services. This can place official information outside normal privacy, security, records and quality controls.
A prohibition may be appropriate for particular tools or information. However, a broad ban without a usable alternative can reduce visibility and leave the operational need unresolved.
A mature NZ government AI response combines clear restrictions with approved accounts, practical training, information-classification rules and a pathway for staff to propose valuable use cases.
Unmanaged adoption
- Personal accounts used for official work
- Sensitive information submitted without assessment
- Outputs accepted without source checking
- No reliable record of prompts, outputs or review
- Use remains invisible to accountable leaders
Governed adoption
- Organisation-managed tools and access
- Clear permitted and prohibited information
- Named use-case owners and human reviewers
- Documented testing and monitoring
- Staff can request support for legitimate needs
Where NZ government AI can create practical value
The most suitable starting points are usually bounded, repeatable and information-heavy workflows where staff retain authority over material decisions.
Lower-impact assistance: retrieve, organise, draft, classify or identify exceptions.
Higher-impact use: determine eligibility, enforcement, entitlement, employment, safety or access to essential services. These uses require substantially stronger evidence and controls.
Moving from an NZ government AI pilot to production
An NZ government AI pilot should answer specific questions about value, feasibility, risk and adoption. It should not become a permanent demonstration disconnected from operational ownership.
| Delivery question | Pilot evidence | Production requirement |
|---|---|---|
| Does it create value? | Measured improvement in time, quality, service or decision support | Benefits case with an accountable service owner |
| Can it perform reliably? | Testing across representative and difficult scenarios | Quality thresholds, monitoring and failure handling |
| Can it be governed? | Defined purpose, permissions, human review and records | Approved controls integrated into normal operations |
| Can people use it? | User feedback, workflow testing and training needs | Adoption plan, support model and updated procedures |
| Can it be sustained? | Indicative cost, integration and support requirements | Funding, ownership, vendor management and lifecycle planning |
How Changeable supports NZ government AI delivery
Changeable helps public organisations translate strategy and guidance into practical, governed implementation.
Frequently asked questions about NZ government AI
How many NZ government AI use cases are operational?
The Government Chief Digital Officer’s 2025 survey reported 55 use cases as deployed and operational across the participating organisations.
How many public-sector AI use cases were reported in 2025?
Seventy government organisations reported a total of 272 AI use cases, including ideas, planned work, trials and operational systems.
Does New Zealand have a national AI strategy?
Yes. New Zealand’s AI Strategy: Investing with confidence was released in July 2025 and focuses on supporting responsible AI adoption and investment.
What guidance applies to public-service generative AI?
Digital.govt.nz provides a Public Service AI Framework and detailed Responsible AI Guidance for the Public Service covering safe, transparent and responsible use.
Can government agencies use AI with personal information?
Potentially, but the Privacy Act and Information Privacy Principles apply. Agencies need an appropriate purpose, safeguards, transparency, authority, governance and usually a Privacy Impact Assessment.
What is a good first NZ government AI use case?
A good starting point is a bounded, repeatable workflow where AI assists with information handling and an authorised person retains responsibility for the final action or decision.
Can Changeable help public organisations implement AI?
Yes. Changeable supports strategy, use-case development, process improvement, governance, document intelligence, workflow automation, AI tools and implementation.
Move NZ government AI from framework to operational value.
Bring us the use case, stalled pilot, workflow or governance question. We will help clarify the outcome, improve the process and design a practical path to controlled implementation.