New Zealand - Public Service AI Framework (RAI-NZ-NA-PUBSEAI-2025)

Overview

The Public Service AI Framework is a GCDO (Government Chief Digital Officer) issued guidance document designed to support responsible AI adoption across New Zealand’s Public Service. It sets a concise vision — "Adopt AI responsibly to modernise public services and deliver better outcomes for all New Zealanders" — and defines five guiding principles that draw on the OECD AI principles. The Framework is presented as non‑binding, risk‑based and proportionate, and sits within New Zealand’s wider AI policy architecture alongside the MBIE work programme and the National AI Strategy. The one‑page A3 summary is available from the GCDO site as a PDF and agencies are encouraged to reference it when creating governance or procurement artifacts. For the official Framework text and the A3 summary see Public Service AI Framework (digital.govt.nz) and the downloadable Public Service Artificial Intelligence Framework (PDF).

Definitions

The Framework adopts a practical working definition of an AI system aligned to the OECD: a machine‑based system that infers from input how to generate outputs (predictions, content, recommendations, decisions) that influence environments. It distinguishes AI by capability (from simple automation to adaptive systems) and covers generative AI as a subset addressed by follow‑on guidance. The Framework also uses standard governance terms such as "human oversight", "assurance", "guardrails", "risk assessment", and "traceability". It instructs agencies to interpret these definitions in light of the agency’s statutory obligations, including privacy, public information and human rights duties. Additional operational definitions and GenAI‑specific considerations are expanded in the companion Responsible AI Guidance for the Public Service: GenAI available at Responsible AI Guidance: GenAI (digital.govt.nz).

Governance and Institutional Framework

The Framework establishes expectations for governance arrangements within agencies and system stewardship by the GCDO. Agencies are expected to implement human accountability at all lifecycle stages, embed risk‑based assurance and reporting, and maintain documentation and audit trails to enable oversight. The GCDO will lead a cross‑agency work programme (Governance, Guardrails, Capability, Innovation, Social licence, Global voice) and coordinate with MBIE, the Public Service Commission, the Office of the Privacy Commissioner and other central agencies. MBIE’s Cabinet paper "Approach to work on Artificial Intelligence" frames the strategic, cross‑portfolio approach and recommends leveraging existing regulatory instruments rather than creating a stand‑alone AI Act; that paper and supporting materials are available at Approach to work on Artificial Intelligence (MBIE PDF). Institutional expectations include escalation mechanisms for high‑risk uses, procurement and supplier assurance processes, and integration of Treaty of Waitangi considerations into governance practices.

Key Focus Areas

The Framework sets out five principles and highlights six pillars that shape the public service’s focus. Principles cover inclusive development, human‑centred values, transparency and explainability, safety and security, and accountability. The pillars (Governance, Guardrails, Capability, Innovation, Social licence, Global voice) direct operational activity: Governance requires clear roles and responsibilities and oversight; Guardrails require risk assessment frameworks, model validation and testing; Capability calls for workforce training and secondments; Innovation supports safe testing and sandboxes for experimentation; Social licence emphasises public engagement, Māori partnership and addressing inequities; Global voice covers international coordination and exportable practices. The Framework also highlights procurement, data ethics, recordkeeping and incident response as recurrent cross‑cutting concerns. For GenAI specific operational measures see the companion guidance and the GCDO pages cited above.

Implementation Framework

The GCDO proposes a work programme and practical tools to help agencies implement the Framework: an assurance model, risk assessment templates, procurement checklists, documentation templates, and capability building resources. Agencies are asked to perform proportionate risk assessments prior to significant AI deployments, to embed security‑by‑design and privacy‑by‑design controls during procurement and development, and to maintain traceable data lineage and documentation to enable audit and redress. Implementation guidance encourages iterative pilots followed by scaled roll‑out when assurance outcomes are acceptable. The Framework deliberately avoids prescriptive thresholds; instead it instructs agencies to size assurance to risk and statutory obligations, using existing legal enforcement channels when breaches occur.

Monitoring and Evaluation

The Framework calls for continuous monitoring, periodic review and reporting to support oversight and public confidence. The GCDO will monitor uptake, maintain a repository of lessons learned and coordinate cross‑agency evaluation. Agencies are expected to maintain documentation sufficient for internal audit, ministerial inquiry and independent review where required. Measurement approaches recommended include post‑deployment performance monitoring, bias and fairness testing, security penetration testing, and user experience assessment. The GCDO will publish updates, and related guidance (e.g., GenAI guidance) supplements monitoring expectations with concrete checklists and reporting templates.

Penalties, Liability, and Appeals

As a non‑binding framework, the document itself does not create statutory penalties. Instead it clarifies that legal liability and enforcement arise under existing laws (Privacy Act 2020, Human Rights Act 1993, Official Information Act 1982, Public Records Act 2005, and civil liabilities) when agencies fail to meet statutory obligations. The Framework emphasises internal accountability, audit, and ministerial oversight as primary levers, while noting that privacy breaches, discriminatory outcomes or unlawful recordkeeping may result in regulatory action, complaints to the Privacy Commissioner, judicial review, or civil liability. Agencies are expected to maintain accessible appeal and review channels for individuals affected by AI‑assisted decisions consistent with administrative law norms.

Relationship to Other Instruments

The Framework is explicitly positioned within New Zealand’s existing legal and policy ecosystem. It references and aligns to the Treaty of Waitangi obligations and domestic statutes (Privacy Act 2020, Public Service Act 2020, Official Information Act 1982, Human Rights Act 1993, Public Records Act 2005). It also references international instruments and best practice, notably the OECD AI Principles. The Framework should be read alongside MBIE’s strategic guidance and the GCDO’s GenAI guidance, and agencies are encouraged to harmonise agency‑level AI policies with the Framework to ensure consistent application across government. See the GCDO page for cross‑references: Artificial intelligence (digital.govt.nz).

International Alignment

The Framework promotes alignment with international norms to support interoperability, trust and New Zealand’s reputation as a trusted AI partner. It endorses OECD values‑based principles, draws lessons from the UK Generative AI work and other international frameworks, and commits to engagement in international fora through MFAT and other agencies. The Global voice pillar directs activity to ensure New Zealand contributes to, and adopts where appropriate, international standards and norms. Links and references include the OECD AI Principles and comparative frameworks referenced on the GCDO page and MBIE materials. The Framework signals a preference for aligning domestic guidance with established international instruments rather than pursuing isolated regulatory regimes.

Implementation Timeline

Sources and References