AI in Public Sector in Aotearoa New Zealand: A Living Whitepaper

Update window: July 14–August 18, 2026

Introduction

AI adoption across New Zealand’s public sector is moving from broad experimentation toward practical, mission-specific deployment. The most recent developments show agencies applying AI to biosecurity standards, tourism information, mental-health service navigation, regulatory work, digital identity, and internal productivity.

The overall model remains cautious: AI is being used to process information, improve search, reduce administrative work, and help staff navigate complex systems. Human experts remain responsible for consequential decisions. However, the policy context is becoming more demanding as the Government links AI with public-service productivity, workforce reform, and the centralisation of digital investment.

Since the July 13, 2026 edition, the key development has been a widening of AI use beyond generic workplace assistance into operational public services and regulatory functions.

Executive Snapshot

• AI adoption is becoming more mission-specific. New initiatives now target biosecurity standards, tourism discovery, mental-health service navigation, and regulator productivity rather than only drafting and summarisation. (beehive.govt.nz)
• The latest quantitative baseline remains the 2025 cross-agency survey: 272 use cases were reported across 70 agencies. The underlying data was self-reported by agencies and was not independently verified. (dia.govt.nz)
• Central coordination is strengthening. The Government Digital Delivery Agency, the Digital Government Target State, and the July Digital Reset Plan are positioning shared digital infrastructure, procurement, and assurance as system-level priorities. (digital.govt.nz)
• Government AI remains primarily assistive. Recent use cases support expert work rather than replacing accountable decision-makers, including the new Biosecurity New Zealand pilot and planned mental-health AI navigation. (beehive.govt.nz)
• Frontier-AI security is now an explicit government concern. The National Cyber Security Centre has advised agencies to prepare for AI-enabled threats by strengthening existing cyber controls and executive accountability. (ncsc.govt.nz)
• Privacy compliance has moved from preparation to implementation. The transition period for existing biometric processing ended on August 3, 2026, while the MBIE Biometric Capability Upgrade has generated further concerns about cost control and financial governance. (privacy.org.nz)
• The public-sector skills gap remains significant. The 2025 State of the Public Service briefing reported that approximately one-third of public servants had tried AI at work, but only 14% used it regularly. (publicservice.govt.nz)
• The central challenge is no longer whether AI can be used, but whether it can be scaled safely, transparently, and cost-effectively.

What Changed Since the July 13, 2026 Edition

1. Biosecurity New Zealand began a focused generative-AI pilot

On July 25, the Government announced a four-month pilot of a generative-AI tool to assist with the development of import health standards. These standards govern the biosecurity requirements for goods entering New Zealand.

The tool is intended to support document-heavy work and improve the speed and consistency of evidence-based standards. Biosecurity New Zealand experts will retain responsibility for key decisions. (beehive.govt.nz)

Why it matters:

• It is a practical example of AI being applied to a highly specialised regulatory workflow.
• The use case combines large-scale document analysis with expert review.
• The pilot directly tests whether AI can reduce administrative burden without weakening biosecurity controls.
• It reflects the Government’s preferred adoption pattern: narrow scope, time-limited testing, and human accountability.

2. Tourism New Zealand received funding to make government-held tourism data “AI-ready”

On July 29, the Government announced an $800,000 investment from the International Visitor Conservation and Tourism Levy to upgrade Tourism New Zealand’s systems for AI search.

The programme will improve the structure of information about tourism and hospitality businesses so that AI tools can more easily discover and accurately present New Zealand travel options. It will also expand Tourism New Zealand’s AI travel assistant with information from regional tourism organisations. (beehive.govt.nz)

Why it matters:

• The public sector is beginning to manage not only the use of AI, but also how government information is represented inside AI systems.
• Data quality, metadata, structured content, and machine readability are becoming public-sector capabilities in their own right.
• This is an example of AI adoption focused on the external information environment rather than internal staff productivity.
• It creates a new public-sector responsibility: ensuring AI-generated recommendations are accurate, current, regionally representative, and culturally appropriate.

3. Mental-health service navigation has entered the formal implementation agenda

The Mental Health and Wellbeing Strategy, launched on August 6, includes a commitment to launch AI Navigation to help people find mental-health and addiction services.

The initiative is framed as a navigation and access tool, intended to help people identify the right support rather than provide autonomous clinical advice. (beehive.govt.nz)

Why it matters:

• It represents a move toward citizen-facing AI in a sensitive service area.
• The proposed use is comparatively bounded: finding and directing people to services.
• The initiative will require careful handling of risk, crisis situations, vulnerable users, accessibility, privacy, and human escalation.
• It reinforces the emerging distinction between AI for service navigation and AI for clinical or eligibility decisions.

4. The biometric compliance deadline has passed

The Biometric Processing Privacy Code 2025 became applicable to existing biometric-processing activities on August 3, 2026. The Code regulates the collection and use of biometric information, including facial, voice, gait, fingerprint, and other behavioural or physiological data. (privacy.org.nz)

The timing is significant because the deadline coincided with renewed scrutiny of MBIE’s Biometric Capability Upgrade project. On July 29, the Immigration Minister said that an additional $6 million in associated costs had been identified and that MBIE could not confirm whether this represented the full expenditure. An independent financial audit was commissioned, and the matter was referred to existing inquiries and the Public Service Commissioner. (beehive.govt.nz)

Implication: biometric processing is becoming a visible test of public-sector technology governance. The issue is not only whether biometric systems are technically effective, but whether agencies can demonstrate necessity, proportionality, transparency, financial control, and responsible stewardship.

5. Digital accessibility is becoming part of the AI operating environment

The Government Digital Delivery Agency consulted on a new Digital Accessibility Standard during July and August 2026. The standard is intended to replace the Web Accessibility Standard in early 2027 and apply accessibility considerations across digital technology, not only websites. (consultations.digital.govt.nz)

This is relevant to AI because public-sector AI systems increasingly influence search, communication, service navigation, and content generation. Accessibility requirements will need to cover:

• AI-generated content and alternative formats.
• Voice, text, and visual interfaces.
• Human fallback channels.
• Accessibility of automated notifications and digital credentials.
• The risk that AI systems reproduce inaccessible or exclusionary content.

Current State of AI Adoption

1. Adoption Scale and Evidence Quality

The latest published cross-agency survey identified 272 AI use cases across 70 agencies, compared with 108 use cases across 37 agencies in 2024. The survey is currently the principal quantitative snapshot of public-sector AI adoption. (dia.govt.nz)

However, newly released information under the Official Information Act clarifies important limitations:

• Use cases were self-reported by agencies.
• Reports were not independently verified.
• Agencies decided how to classify and describe their own use cases.
• The Government Chief Digital Officer undertook data cleansing for duplicates and formatting, but not full external validation.
• The survey is intended to identify adoption patterns, benefits, shared opportunities, and common barriers. (dia.govt.nz)

Assessment: the survey is valuable for understanding direction and breadth, but it should not be treated as a precise measure of production maturity, financial return, or service impact.

2. Governance and Central Coordination

The public-sector AI governance model is now distributed across several layers:

• The Public Service AI Framework.
• Responsible AI guidance for generative AI.
• The Public Service AI Work Programme.
• The Government Digital Delivery Agency.
• Sector-specific guidance, including regulator and health guidance.
• Privacy, security, records-management, and information-sharing obligations.

The Government Digital Delivery Agency was established on April 1, 2026, within the Public Service Commission. It inherited the functions of the former Government Chief Digital Office and is intended to provide stronger system leadership for digital delivery, capability, investment, procurement, and common platforms. (digital.govt.nz)

The July Digital Reset Plan found that digital investment remained fragmented, with duplication, weak system-level prioritisation, and limited central influence over funding, design, and procurement. It recommended a more coordinated model for foundational digital capabilities. (publicservice.govt.nz)

Emerging operating model:

Shared digital and AI infrastructure at the centre, with agency-specific applications at the edge.

This model is visible in the planned AI broker or gateway, AI platform services, semantic search, shared identity infrastructure, and the Govt.nz app. It is also consistent with the OECD’s assessment that the GDDA is centralising leadership for digital investment, procurement, and delivery. (oecd.org)

3. Workforce Adoption

The 2025 State of the Public Service briefing provides the latest system-level workforce picture:

• Around one-third of public servants had tried AI at work.
• Only 14% used AI regularly.
• Most staff expressed confidence in learning new digital skills.
• Adoption remained uneven between agencies. (publicservice.govt.nz)

The gap between experimentation and regular use is strategically important. It suggests that many public servants have access to or awareness of AI tools, but institutional adoption is constrained by:

• Unclear business processes.
• Limited training.
• Privacy and security concerns.
• Uncertainty about acceptable use.
• Lack of reliable measurement.
• Concerns about accuracy and accountability.
• Inconsistent access to approved tools.

The Government’s AI adoption challenge is therefore less about introducing tools and more about redesigning work around them.

4. Security and Resilience

The National Cyber Security Centre’s June 2026 guidance places frontier AI within the government cyber-risk environment. It warns that advanced AI can increase both defensive capability and the speed, scale, and affordability of malicious cyber activity. (ncsc.govt.nz)

The guidance recommends that government agencies:

• Confirm executive accountability for frontier-AI cyber risk.
• Review compliance with the New Zealand Information Security Manual and Protective Security Requirements.
• Identify material vulnerabilities that AI-enabled attackers could exploit.
• Maintain strong identity, access, patching, monitoring, recovery, and incident-response controls.
• Avoid assuming that access to the most advanced AI models is necessary for effective cyber readiness. (ncsc.govt.nz)

Key insight: frontier AI is increasing the importance of basic cyber hygiene rather than replacing it. Agencies adopting AI without strong identity, data, logging, and supplier controls will face amplified risks.

5. Public-Facing Digital Services

The Govt.nz app remains the clearest example of the Government’s shared digital-service architecture. It includes a digital wallet, access to government information and services, and an infrastructure for digital credentials. The programme also anticipates secure messaging, notifications, and further agency integration. (digital.govt.nz)

While the app is not itself an AI system, it provides an important platform for future AI-enabled service navigation and personalised interactions. Its significance lies in the shared channel and identity infrastructure that AI services may eventually use.

The 2025 AI assistant pilot also found strong demand for simpler government navigation: 85% of participants said the AI assistant was more efficient than their previous methods for finding government information and services. (digital.govt.nz)

The main design principle remains that digital assistance must not eliminate non-digital access. The Govt.nz app is explicitly optional, and agencies remain responsible for providing services through multiple channels. (digital.govt.nz)

Recent News and Policy Developments

AI is being tied more directly to productivity reform

The Government’s broader public-service reform programme continues to connect workforce productivity with increased use of AI and digital tools. This has created political and analytical debate about whether AI is being treated as a genuine productivity capability or as an assumption supporting staff reductions.

The University of Auckland’s Professor Alexandra Andhov has argued that the Government has not published sufficiently detailed estimates of AI’s total cost, including licences, model usage, implementation, oversight, audit, error correction, and ongoing human review. (auckland.ac.nz)

This critique identifies a key weakness in current public-sector AI debate: salary savings are easier to quantify than the full cost of reliable and accountable automation.

Government is investing in AI capability outside the public service

The Government launched new secondary-school subjects in August, including Applied Intelligent Systems. The subject includes low- and no-code technologies, AI-enabled workflows, and autonomous agents. (beehive.govt.nz)

Although this is primarily an education and workforce measure, it is relevant to government adoption because public agencies will need a larger pool of people able to:

• Evaluate AI vendors.
• Design safe workflows.
• Manage data and model risk.
• Monitor outputs.
• Integrate AI into existing services.
• Work across policy, technology, legal, and operational functions.

New Zealand’s wider digital-government performance remains uneven

The OECD Digital Government Outlook 2026 recognises that AI is already used across multiple areas of New Zealand government and notes the creation of the GDDA. It also identifies continuing weaknesses in open data, user-driven service design, impact evaluation, ex-post cost-benefit analysis, GovTech strategy, and standardised project management. (oecd.org)

These weaknesses matter directly to AI adoption. Poor data availability, weak evaluation, and inconsistent service metrics make it harder to determine whether AI systems improve outcomes or simply add another layer of technology.

Research Overview

1. 2025 Cross-Agency AI Survey

Main findings:

• 272 reported use cases.
• 70 participating agencies with reported use cases.
• More than twice the number of use cases recorded in 2024.
• Use cases span productivity, service delivery, analytics, search, automation, and security.

Important qualification: the use cases were self-reported and not independently verified. (dia.govt.nz)

2. State of the Public Service 2025

The Public Service Commission’s three-yearly briefing presents AI as a major enabler of productivity, improved customer experience, and more responsive government. It identifies use cases including:

• Tax administration.
• Biosecurity risk detection.
• Public-facing chatbots.
• Summarisation and drafting.
• Theme identification.
• Information retrieval.
• Speech-note reduction.
• Service navigation.

The briefing also stresses that AI can introduce bias, lack contextual nuance, and require human oversight to preserve public trust. (publicservice.govt.nz)

3. OECD Digital Government Outlook 2026

The OECD’s assessment places New Zealand’s AI adoption within a broader digital-government context. It recognises that AI is being applied across government but highlights gaps in:

• Open and reusable data.
• User-driven service design.
• Impact measurement.
• Digital investment evaluation.
• Whole-of-government delivery capability.
• Standardised service metrics.

The report supports the conclusion that New Zealand has made progress on institutional architecture but still needs stronger implementation and evaluation capability. (oecd.org)

4. Public-sector AI cost and governance analysis

University of Auckland analysis has challenged the assumption that AI automatically reduces public-sector costs. It argues that government must account for:

• Recurring model and licence costs.
• Integration and infrastructure.
• Human review.
• Procurement expertise.
• Auditing and assurance.
• Error correction.
• Vendor dependence.
• Data sovereignty and offshore expenditure.

This analysis is not an official Government position, but it is an important counterweight to productivity-focused policy messaging. (auckland.ac.nz)

Case Studies

Case Study 1: Biosecurity New Zealand’s AI-assisted import health standards

Description

Biosecurity New Zealand is developing and testing a generative-AI tool to assist with import health standards. A four-month pilot is evaluating whether the tool can support document analysis and produce high-quality, evidence-based material more efficiently. (beehive.govt.nz)

Adoption model

• Narrow, document-heavy workflow.
• Expert review retained.
• Time-limited pilot.
• Evidence-based output requirement.
• No autonomous final decision-making.

Strategic significance

This is one of the strongest current examples of AI being applied to a specialised regulatory function. It demonstrates a pattern likely to be repeated across government: AI will first enter high-volume knowledge work where outputs can be reviewed by domain experts.

Case Study 2: Tourism New Zealand’s AI-ready information systems

Description

Tourism New Zealand is upgrading its systems so tourism products and services are more discoverable through AI search. The programme includes structured business information, improved data quality, and expansion of an AI travel assistant using regional tourism content. (beehive.govt.nz)

Strategic significance

This case shows that public-sector AI adoption is not limited to building chatbots or buying models. Agencies must also prepare their information assets for machine-mediated discovery.

The project raises important governance questions:

• Who verifies the accuracy of AI-accessible information?
• How are small and regional operators represented?
• How are Māori tourism experiences described?
• How are commercial interests balanced with public information responsibilities?
• How are errors corrected when AI systems reproduce outdated information?

Case Study 3: Public Service Commission use of Microsoft Copilot

Description

An Official Information Act release reported that Microsoft Copilot was rolled out to Public Service Commission staff in August 2025. During October–December 2025:

• 209 users engaged with AI tools.
• 19,224 prompts were submitted.
• Average use was approximately 470 prompts per day.
• The Commission’s Digital Services team held responsibility for governance and systems management. (publicservice.govt.nz)

Strategic significance

This provides one of the clearest publicly documented examples of routine generative-AI use inside a central government agency.

It also illustrates the difference between:

• Tool adoption, measured by access and prompts.
• Effective adoption, measured by time saved, quality improved, risk reduced, or services enhanced.

The public release provides usage data but does not establish a quantified return on investment. That measurement gap is common across the sector.

Case Study 4: AI Navigation for mental-health and addiction services

Description

The Mental Health and Wellbeing Strategy includes a commitment to launch AI Navigation to help people locate appropriate mental-health and addiction services. (beehive.govt.nz)

Strategic significance

This is a potentially high-value but high-sensitivity use case. Successful deployment will depend on:

• Clear boundaries between navigation and clinical advice.
• Rapid escalation for crisis situations.
• Human alternatives.
• Accessibility and culturally safe design.
• Protection of sensitive personal information.
• Testing with people who have lived experience.
• Ongoing monitoring for harmful or misleading recommendations.

The initiative is currently best understood as a planned service capability rather than evidence of a fully operational national AI system.

Case Study 5: Biometric Capability Upgrade and governance risk

Description

MBIE’s seven-year Biometric Capability Upgrade project was ceased in December 2025. In July 2026, the Government disclosed that an additional $6 million in associated costs had been identified and that a further audit was required because the full expenditure could not yet be confirmed. (beehive.govt.nz)

Strategic significance

The project is not a conventional AI adoption success story. Its relevance lies in the governance lessons:

• Biometric and AI projects can be difficult to cost accurately.
• Technology projects involving identity data require strong assurance.
• Financial controls and technical governance are inseparable.
• Public trust can be damaged by project failure even before a system is deployed.
• The Biometric Processing Privacy Code creates enforceable obligations around necessity, proportionality, transparency, accuracy, security, and use limits. (privacy.org.nz)

Adoption Trends

1. From generic productivity to mission-specific AI

Earlier public-sector adoption was concentrated in summarisation, drafting, search, transcription, and internal workflow support. Newer initiatives are increasingly tied to defined policy and service problems:

• Biosecurity standards.
• Tourism discovery.
• Mental-health service access.
• Regulatory analysis.
• Digital identity and public-service navigation.

This is a positive maturation signal because narrowly defined problems are easier to evaluate and govern.

2. Human-in-the-loop remains the dominant model

Across current initiatives, people remain responsible for:

• Regulatory decisions.
• Clinical judgements.
• Service eligibility.
• Public-facing accountability.
• Risk acceptance.
• Quality assurance.

This approach is reflected in regulator guidance, Biosecurity New Zealand’s pilot, and the planned mental-health navigation service. (regulation.govt.nz)

3. Data readiness is becoming as important as model capability

Tourism New Zealand’s investment demonstrates that AI adoption depends on structured, accurate, current, and machine-readable information. The OECD’s concerns about New Zealand’s open-data performance reinforce the same point. (beehive.govt.nz)

4. Centralised infrastructure is increasingly preferred

The GDDA, Digital Government Target State, Govt.nz app, shared digital credentials, and digital reset process all point toward common platforms and shared capabilities rather than isolated agency-by-agency systems. (digital.govt.nz)

5. AI adoption is increasingly connected to workforce redesign

The Government is simultaneously:

• Encouraging AI adoption.
• Reducing public-service costs.
• Restructuring agencies.
• Introducing AI-related education and training.
• Building shared digital infrastructure.

This creates a risk that AI becomes associated primarily with job reduction rather than service improvement. The long-term success of adoption will depend on whether agencies can demonstrate better outcomes for the public, not simply lower headcount.

Risks and Pressure Points

Skills and capability

Regular use remains low relative to general awareness. Agencies need more capability in:

• AI procurement.
• Data governance.
• Model evaluation.
• Records management.
• Privacy and security.
• Workflow redesign.
• Algorithmic assurance.
• Māori data governance and sovereignty.

Cost transparency

Public-sector AI business cases should include the total cost of ownership:

• Model and licence fees.
• Cloud and data costs.
• Integration.
• Staff training.
• Human checking.
• Assurance and auditing.
• Security controls.
• Vendor switching or exit costs.
• Accessibility and support.

The University of Auckland analysis highlights the danger of counting salary reductions while excluding recurring technology and oversight costs. (auckland.ac.nz)

Data sovereignty and cultural risk

AI systems may process information offshore, reproduce biased representations, or use Māori data without sufficient attention to tikanga, Māori interests, or Māori data sovereignty.

These risks are particularly significant in:

• Health.
• Social services.
• Identity.
• Biometrics.
• Justice.
• Environmental and cultural data.
• Public-facing information systems.

Public trust and transparency

The Government’s general trust base remains stronger than public understanding of government AI. The sector therefore faces a transparency challenge: people need to know when AI is being used, what role it plays, what information it uses, and how to reach a human decision-maker.

Cybersecurity

AI-enabled threats may increase the speed and scale of phishing, vulnerability discovery, fraud, impersonation, and automated attacks. Agencies need to treat AI as both a technology capability and a change in the threat environment. (ncsc.govt.nz)

Accessibility and exclusion

AI systems may disadvantage people who:

• Have disabilities.
• Prefer non-digital channels.
• Have limited digital literacy.
• Speak languages underrepresented in training data.
• Require culturally specific support.
• Need human assistance in complex or stressful situations.

Outlook: August 2026–2027

The next phase of New Zealand public-sector AI adoption is likely to be defined by five priorities.

1. Scaling proven workflow use cases

Agencies will continue to expand AI for:

• Document analysis.
• Search and retrieval.
• Transcription.
• Case preparation.
• Regulatory evidence review.
• Contact-centre assistance.
• Service navigation.

2. Building shared AI infrastructure

The Government is likely to place greater emphasis on:

• Shared AI gateways and brokers.
• Common procurement.
• Approved model access.
• Security controls.
• Semantic search.
• Shared evaluation tools.
• Common records and audit requirements.

3. Measuring benefits more rigorously

Future reporting will need to distinguish among:

• Number of users.
• Number of prompts.
• Number of use cases.
• Production deployment.
• Time saved.
• Cost avoided.
• Service quality.
• Customer outcomes.
• Error and escalation rates.
• Equity and accessibility outcomes.

4. Establishing stronger assurance for high-impact uses

Biometrics, health, welfare, identity, justice, immigration, and public safety will require more formal assurance than low-risk drafting or summarisation tools.

5. Managing the relationship between AI and public-service employment

AI may reduce repetitive administrative work, but it will also create demand for new roles in:

• AI assurance.
• Data stewardship.
• Digital service design.
• Model monitoring.
• Privacy and security.
• Procurement.
• Human-centred service support.

The central policy question will be whether AI is used to strengthen frontline capacity and service quality, or primarily to reduce staffing without sufficient investment in implementation capability.

Overall Assessment

As of August 18, 2026, AI adoption in New Zealand’s public sector is best described as structured expansion under fiscal and governance pressure.

The sector is no longer operating only through isolated experiments. AI is now being introduced into defined government missions, including biosecurity, tourism, mental-health navigation, regulatory work, and internal public-service operations. Shared digital infrastructure and central coordination are also becoming more prominent through the GDDA and the Digital Government Target State.

However, the evidence base remains incomplete. The latest cross-agency survey shows rapid growth in reported use cases, but the data is self-reported and not independently verified. Publicly available information often describes pilots, intentions, or tool usage rather than measured improvements in outcomes.

The most credible adoption pattern remains bounded augmentation:

• AI processes information.
• AI supports staff.
• AI improves search and navigation.
• AI reduces administrative effort.
• Humans remain accountable for consequential decisions.

The main test for the next year will be whether New Zealand can convert this cautious adoption model into reliable, measurable, and trusted public value. That will require stronger cost accounting, better impact evaluation, robust cyber and privacy controls, transparent public communication, and sufficient internal capability to govern systems that are increasingly central to public-service delivery.