New Zealand Public Service AI Framework

What it is

The Public Service AI Framework is the New Zealand Government’s framework for the responsible use of AI across the public service. It was published by the Government Chief Digital Officer (GCDO), part of the Department of Internal Affairs, and last updated on 29 January 2025.

Its vision is to “adopt AI responsibly to modernise public services and deliver better outcomes for all New Zealanders”. It sets out principles for how AI should be used and the pillars of the GCDO’s work programme for getting there. It now sits within New Zealand’s AI Strategy, launched in July 2025, and alongside the GCDO’s Responsible AI Guidance for the Public Service on generative AI.

Who it applies to

Public service agencies. It is aimed at leaders, decision-makers, practitioners and others who influence AI in public service agencies, though it may be useful more widely. It applies to all forms of AI used in New Zealand public services.

The framework is not binding. Agencies are encouraged to align with the direction it sets. Legal obligations under the Privacy Act 2020 and other legislation apply to AI use regardless.

The five principles

The framework’s principles closely follow the values-based OECD AI Principles.

1. Inclusive, sustainable development
2. Human-centred values
3. Transparency and explainability
4. Safety and security
5. Accountability

The principles describe how AI should be used, not a procedure. They are the test an agency’s own AI policy, risk assessments and transparency material should be able to answer to.

The six pillars

The pillars set the focus of the GCDO’s Public Service AI work programme.

• Governance. Supporting transparency and human accountability in public service AI use.
• Guardrails. Enabling safe and responsible public service AI use.
• Capability. Building internal and external AI knowledge and skills.
• Innovation. Providing pathways for safe AI testing and innovation in the public service.
• Social licence. Making sure New Zealanders and public service workers have trust and confidence in public service AI use.
• Global voice. Making sure international counterparts see New Zealand as a trusted AI partner.

What it means for an agency

The framework is directional, so an agency turns it into practice through its own policies. Read against the principles and pillars, a sound agency approach has a few recognisable parts.

• Named accountability. A person accountable for each AI use, which is what the governance pillar and the accountability principle rely on.
• A record of AI in use. Agencies cannot be transparent about AI they have not found. An internal list of AI systems and tools, including those staff adopted themselves, is the starting point.
• Assessment before deployment. Risk and privacy impact assessment in proportion to the use, which is what the guardrails pillar and the safety and security principle point to.
• Openness. Plain information for the public about where AI is used and how, which supports the social licence pillar.
• Capability. Training for staff who use or oversee AI, matching the capability pillar.
• Safe experimentation. A defined route for trials, so innovation happens inside guardrails rather than around them. Agencies that make algorithmic decisions about people also have the Algorithm Charter, which many have signed. Any agency handling personal information with AI has to meet the Information Privacy Principles, as the Privacy Commissioner sets out.

The generative AI guidance

The framework is the umbrella. For generative AI specifically, the GCDO publishes the Responsible AI Guidance for the Public Service: GenAI, a collection of guidance on using generative AI tools safely in agencies. Other government bodies apply it to their own work. MBIE, for example, says it used AI to develop parts of its business guidance in line with the GCDO’s GenAI guidance, with human oversight of anything generated. An agency’s own generative AI rules, such as which tools are approved and what information may be entered, sit underneath both.

Key dates

The framework was last updated on 29 January 2025. New Zealand’s AI Strategy, which the framework now sits within, was launched in July 2025, along with MBIE’s Responsible AI Guidance for Businesses.

How it relates to other frameworks

The five principles track the OECD AI Principles, to which New Zealand adheres. The Algorithm Charter adds specific commitments and a risk matrix for algorithms that affect people, and the Privacy Act 2020 supplies the legal baseline. For Australian agencies, the closest counterparts are the DTA policy and the national framework for the assurance of AI in government, both of which carry more specific requirements.