AI in Public Sector in Aotearoa New Zealand: A Living Whitepaper
Updated: April 2, 2026
Introduction
AI adoption in New Zealand’s public sector has moved another step forward since the previous March 9, 2026 update. The core story is no longer just policy formation or isolated pilots. It is now about execution: shared architecture, all-of-government channels, production deployments in bounded service contexts, more visible public disclosures, and tighter governance around who can use AI, for what, and under what controls. At the same time, adoption is still uneven, frontline use remains carefully scoped, and trust is still the limiting variable. (digital.govt.nz)
Executive Snapshot
- Public-sector AI is now operating at portfolio scale. The 2025 cross-agency AI survey found 70 agencies reporting 272 use cases, including 55 operational deployments, up from 108 use cases across 37 agencies in 2024. Most activity is still in internal productivity, analysis, search, and workflow support rather than autonomous frontline decision-making. (digital.govt.nz)
- The next phase is infrastructure, not just experiments. The January 2026 Public Service AI Work Programme and the February 2026 Digital Target State show central government designing for shared AI capabilities such as AI platform services, an AI broker/gateway, semantic search, data and AI safeguards, and all-of-government channels including the Govt.nz app. (digital.govt.nz)
- Customer-facing AI is expanding, but in tightly bounded forms. The GCDO’s Govt.nz AI Assistant is planned as a public tool and as part of the Govt.nz app, while agencies such as NZQA and Health New Zealand are using AI in scoped service contexts such as public website support and clinical documentation. (digital.govt.nz)
- Health is now one of the clearest examples of scaled frontline use. As of February 28, 2026, every emergency department in New Zealand had access to an AI scribe tool, with rollout completed to 1,250 ED doctors and frontline staff. The government said pilot results showed doctors using the tool saw one additional patient per shift on average. (beehive.govt.nz)
- Capability remains a bottleneck. The 2025 Public Service Census found 33% of public servants had tried AI at work, but only 14% used it regularly; only 42% agreed their organisation takes advantage of technology to deliver better services. (publicservice.govt.nz)
- Trust and transparency are still the key constraints. Kantar found only 4% of New Zealanders felt well informed about how AI is used by government agencies; 55% said they were comfortable with government use of AI, while 45% remained uneasy. (kantarnewzealand.com)
What Changed Since the March 9, 2026 Edition
1) The “shared digital government” layer is now much clearer
The biggest strategic change is that AI is being embedded into the government’s broader digital operating model. The February 2026 Digital Target State explicitly places AI Platform Services, an AI Broker / Gateway, Semantic Search, and Data & AI Safeguards inside a new digital public infrastructure layer. This is a stronger and more concrete articulation than the earlier whitepaper’s focus on general coordination. (dns.govt.nz)
2) The Govt.nz app has moved from roadmap to live channel
As of early April 2026, the Govt.nz app is live in its second release with a digital wallet setup feature, emergency alerts, and a plan to add more integrated services through 2026. The app page says the wallet is available from early April 2026, with the first accredited digital credentials expected from June 2026. While the AI assistant is not presented there as already live in-app, the broader work programme confirms the Govt.nz AI Assistant is intended to sit inside the app. (govt.nz)
3) A major frontline deployment arrived in health
The previous edition focused mostly on back-office and service-navigation use. The February 2026 nationwide rollout of AI scribe across emergency departments materially changes the picture by showing New Zealand can move from pilot to national operational deployment in a clinical workflow, albeit one that supports documentation rather than diagnosis or fully automated decision-making. (beehive.govt.nz)
4) Transparency is improving through public disclosure, not just policy
Two concrete examples stand out. DPMC’s Honours Unit now publicly states it may use a generative AI tool to summarise nomination material and support letters, with human checking and no retention or external sharing of personal information. And NZQA published new OIA releases on March 6, 2026 covering AI marking and assessment and May 2025 automated text scoring results. (dpmc.govt.nz)
Current State of AI Adoption
Governance and Operating Model
New Zealand’s public-service AI governance stack remains anchored in the Public Service AI Framework and the Responsible AI Guidance for the Public Service: GenAI. The framework’s five principles are inclusive and sustainable development, human-centred values, transparency and explainability, safety and security, and accountability, and it explicitly says agencies should disclose when AI is used, how it was developed, and how it affects outcomes where relevant. (digital.govt.nz)
Since January 2026, that governance stack has been complemented by the Public Service AI Work Programme to 2027, which shifts the centre of gravity from policy-setting to implementation. The programme includes initiatives around a Govt.nz AI Assistant, international collaboration, shared tools, and reusable standards and patterns. (digital.govt.nz)
Scale and Maturity of Use
The best available system-wide evidence still comes from the 2025 GCDO survey. It shows a jump to 272 use cases across 70 agencies, including 207 use cases from 44 public service organisations and 65 use cases from 26 wider public sector organisations. Operational use is growing: 55 use cases were reported as deployed and in operation, compared with 15 a year earlier. (digital.govt.nz)
The same survey suggests the sector’s maturity is improving in two important ways:
- more agencies are participating
- more use cases are reaching operational stage
- more organisations are adapting and reusing tools rather than starting from scratch. (digital.govt.nz)
Workforce Readiness
Workforce adoption is real but still shallow. The 2025 Public Service Census found 33% of public servants had ever used AI for work and 14% used it regularly. Although 88% said they felt confident learning new digital skills, fewer than half said their organisation was taking advantage of technology to improve public services. (publicservice.govt.nz)
That aligns with the State of the Public Service 2025, which says the public service still suffers from slow uptake of enabling technologies such as AI and a fragmented operating model, even as the system begins coordinated acceleration. (publicservice.govt.nz)
Recent News and Policy Developments
Shared AI architecture is now part of the digital-government blueprint
The February 2026 Digital Target State is one of the most important developments since the last update because it makes AI part of the future architecture of government service delivery rather than a standalone innovation stream. It describes a standardised digital public infrastructure layer with common AI tooling, privacy-preserving service orchestration, central safeguards, and semantic search across government content and services. (dns.govt.nz)
The Govt.nz app is becoming the all-of-government front door
The Service Modernisation Roadmap V2 says the GCDO will deliver the Government App and an AI virtual assistant to help people find and use government information and services confidently. The live Govt.nz app now offers a practical channel for that strategy, with digital wallet setup, emergency updates, and further integrated services planned through 2026. (digital.govt.nz)
DPMC has disclosed a live GenAI use case in the honours system
A notable operational detail now publicly confirmed is that the Honours Unit may use a generative AI tool to summarise nomination material and support letters. DPMC states that human review is applied, the tool does not retain personal information, and data is not shared outside DPMC. That is significant because it shows central agencies using GenAI in real administrative workflows under explicit privacy conditions. (dpmc.govt.nz)
Health New Zealand has moved beyond pilots in a frontline setting
Health Minister Simeon Brown announced on February 28, 2026 that AI scribe technology was live in all emergency departments, following a successful pilot. This is one of the clearest examples in New Zealand of AI moving into a national frontline workflow, but it remains an assistive use case focused on documentation rather than clinical autonomy. Health New Zealand’s governance model reinforces that caution: proposals to develop or implement AI tools must be registered with its National Artificial Intelligence and Algorithm Expert Advisory Group. (beehive.govt.nz)
Research and Evidence Base
Cross-agency evidence
The GCDO survey remains the strongest quantitative view of adoption. Its headline message is simple: AI use in government has moved from a scattered set of experiments to a recognisable portfolio with live deployments, planning pipelines, and visible reuse patterns. The survey also reports that current use is concentrated in supporting functions such as data, administration, communications and strategy, with only a small percentage in direct customer-facing service delivery. (digital.govt.nz)
System-level assessment
The State of the Public Service 2025 adds an institutional reading to those numbers. It points to AI already making a difference in areas such as tax administration, biosecurity risk detection, and public-facing chatbots, but says New Zealand remains below leading jurisdictions in AI readiness and has been slower to implement AI systematically. It also cites the GCDO assistant pilot, noting 85% of participants found the tool more efficient than previous search methods. (publicservice.govt.nz)
Public trust and awareness
Public trust is now a practical delivery issue. Kantar’s late-2025 New Zealand research found very low public understanding of government AI use, with only 4% saying they felt well informed. Comfort levels were mixed rather than overwhelmingly positive. This matters because the public sector’s ability to expand AI in higher-stakes areas will depend not only on safety controls but on whether agencies can explain what is being used, where, and why. (kantarnewzealand.com)
Case Studies
Case Study 1: Govt.nz AI Assistant and the shift to all-of-government service navigation
The GCDO’s Govt.nz AI Assistant remains the clearest central-government example of bounded customer-facing AI. The work programme describes it as an AI search assistant public tool for government information that will also be available in the Govt.nz app. Combined with the new digital target state and app rollout, this suggests the government is building toward a shared conversational service-navigation layer rather than dozens of fragmented agency assistants. (digital.govt.nz)
Why it matters: this is the strongest signal that New Zealand’s next phase is platformisation. The emphasis is on safe search, orchestration, and service discovery, not open-ended autonomous government. (dns.govt.nz)
Case Study 2: Health New Zealand’s AI scribe rollout
The nationwide ED rollout marks one of the most advanced public-sector AI deployments in the country. It is operational, national in scale, and tied to measurable workflow value: the government says doctors using the tool saw one more patient per shift on average during the pilot. At the same time, Health NZ maintains formal AI review machinery through its expert advisory group and proposal process. (beehive.govt.nz)
Why it matters: this shows New Zealand is willing to deploy AI in frontline public services when the use case is tightly defined, human-supervised, and operationally measurable. (beehive.govt.nz)
Case Study 3: NZQA’s AI-enabled assessment and service operations
NZQA remains one of the most mature agency adopters. It has used AI-powered automated text scoring for large-scale literacy assessment marking, and its website chatbot Awhina has been upgraded with retrieval-style GenAI functionality. Importantly, NZQA also published new OIA releases in March 2026 covering AI marking and assessment and automated text scoring results, which is a meaningful transparency signal. (www2.nzqa.govt.nz)
Why it matters: NZQA demonstrates that consequential use is possible in New Zealand when models are tightly scoped, tested, and paired with strong quality controls and public explanation. (www2.nzqa.govt.nz)
Case Study 4: Hutt City Council’s AI-Volution programme
Hutt City Council remains the most advanced visible local-government example. Its current public AI page says it has 300 secure AI licences, a public AI Register, an AI Risk Management Framework, an internal governance group, and a growing set of custom assistants covering meeting minutes, project briefs, traffic management review, building consent support, consultation analysis, and internal communications. It also says the council is trialling AI for LIM processing and resource-consent workflows, with invoicing automation saving 3 to 5 minutes per invoice. (huttcity.govt.nz)
The earlier 2025 award announcement remains relevant because it quantified the programme at 15 custom-built assistants, 44,000 staff hours reclaimed annually, and around NZ$900,000 in annual savings. (huttcity.govt.nz)
Why it matters: Hutt shows what mature public-sector GenAI adoption looks like when an organisation is willing to combine strategy, policy, training, governance, tooling, and public disclosure. (huttcity.govt.nz)
Case Study 5: DPMC’s Honours Unit
The Honours Unit’s disclosed use of GenAI to summarise nominations is a smaller use case, but strategically important. It is a real central-agency administrative deployment with privacy language, human review, and clear limits on retention and external sharing. (dpmc.govt.nz)
Why it matters: it illustrates the dominant public-service pattern in 2026: use AI first where it can compress administrative effort without removing accountable human judgment. (dpmc.govt.nz)
Adoption Patterns and Trends
1) New Zealand has entered the implementation era
The evidence now points to a sector moving from “pilot and policy” to “platform and production.” Central leadership is no longer just issuing guidance; it is defining shared channels, common AI services, and reusable safeguards. (digital.govt.nz)
2) The dominant model is still augmentation, not replacement
The bulk of current use remains assistive: summarisation, search, workflow acceleration, document drafting, thematic analysis, public navigation, and clinical or administrative support. Even where use is consequential, the pattern is still human-in-the-loop rather than fully automated state action. (digital.govt.nz)
3) Shared infrastructure is becoming the main strategic lever
The emergence of AI platform services, AI broker/gateway, and semantic search in the digital target state suggests that the next productivity gains will come less from isolated pilots and more from common components that multiple agencies can consume. That is a strong sign of institutional maturation. (dns.govt.nz)
4) Transparency is slowly moving from principle to practice
Public registers, public statements, OIA releases, and explicit app disclosures are becoming more common. Hutt City Council’s AI Register, DPMC’s honours AI statement, and NZQA’s March 2026 AI OIA releases are all examples of this shift. But the Kantar data suggests the public still does not feel well informed overall. (huttcity.govt.nz)
5) Local government is becoming an important innovation layer
The most operationally ambitious public examples are not all from Wellington. Hutt City Council’s programme shows local government can move quickly where the value case is strong and the use cases are well bounded. Invercargill City Council’s 2025–2026 committee material also shows another council formalising an AI use policy, approved-tools list, and acknowledgement requirements. (huttcity.govt.nz)
Pressure Points and Risks
Skills, cost, security, and privacy remain the core barriers
The cross-agency survey still points to capability gaps, anticipated cost, and security concerns as major barriers. That has not changed; if anything, it becomes more important as agencies move from experimentation to operationalisation. (digital.govt.nz)
Governance failures are now an operational risk
The Department of Corrections story in February 2026 is a reminder that AI misuse is not hypothetical. RNZ reported staff were warned after some had used AI to draft formal reports outside approved use. That reinforces a broader lesson: once tools become commonplace, the challenge is no longer just whether to adopt AI, but how to control everyday use inside normal public-service workflows. (rnz.co.nz)
Biometrics will be a major compliance test in 2026
The Biometric Processing Privacy Code 2025 came into force on November 3, 2025, with the transition period for existing biometric processing ending on August 3, 2026. That deadline will matter for agencies using or considering facial recognition and other biometric tools. (privacy.org.nz)
Overall Assessment
AI adoption in New Zealand’s public sector is now best described as managed scaling. The sector has moved beyond fragmented experimentation: it has a system-wide framework, formal responsible-use guidance, a central work programme, shared architectural thinking, and a growing set of operational use cases across central and local government. (digital.govt.nz)
But the current model is still deliberately conservative. New Zealand is expanding AI mostly where it can improve productivity, navigation, documentation, and bounded decision support without removing accountable human oversight. That is visible in the honours system, council operations, health documentation, public search, and NZQA’s tightly scoped assessment tooling. (dpmc.govt.nz)
The strategic implication is clear: the public sector is no longer asking whether it will use AI, but how to industrialise it safely. The winners over the next 12 months will be the agencies that can combine reusable infrastructure, workforce capability, public transparency, and measurable service improvement without losing trust. As of April 2, 2026, New Zealand has made visible progress on all four dimensions, but trust and institutional capability remain the hard limits on how fast adoption can go. (publicservice.govt.nz)