New Zealand - Algorithm Use Guidelines

Overview

The Algorithm Charter for Aotearoa New Zealand (July 2020) is a voluntary, cross-government commitment designed to increase public confidence in the way public agencies develop and use algorithms. Rather than a prescriptive technical standard, the Charter uses a risk‑matrix approach to ensure that agencies focus attention and resources on algorithmic uses that have a moderate to high likelihood of unintended adverse outcomes or would have significant impact if they go wrong. The Charter sets out six core commitments — Transparency; Partnership (including embedding Te Ao Māori perspectives and Treaty of Waitangi principles); People (engagement and consultation); Data (fit for purpose and bias mitigation); Privacy, Ethics and Human Rights (peer review and safeguards); and Human Oversight (nominated contacts and appeal channels). The Charter is published on the government data toolkit and accompanied by practical tools (threshold assessments, impact assessment questionnaires, and guidance) and an implementation plan that phases guidance, transparency, and oversight measures. The original publication and subsequent implementation plan are available from the government data toolkit and related public sector pages such as the Beehive launch statement.

Definitions

The Charter intentionally does not adopt a single technical definition of an "algorithm". It recognises a spectrum of techniques — from simple decision rules and regression models to machine learning approaches such as neural networks and Bayesian models. Instead of a strict definitional boundary, the Charter focuses on the effects and risks of algorithm use: agencies must identify algorithmic deployments that could significantly affect wellbeing or have a high likelihood of causing unintended harm, and apply the Charter commitments in proportion to assessed risk. The Charter references prior analysis such as "Government Use of Artificial Intelligence in New Zealand" (2019) as background for definition and classification discussions.

Governance and Institutional Framework

The Charter is a non‑statutory framework stewarded within the New Zealand government data system and promoted via the central data toolkit hosted on data.govt.nz. Leadership for uptake, review and implementation is coordinated through system leads including Tatauranga Aotearoa — Stats NZ, with implementation support from a cross‑agency community of practice. The implementation plan contemplates three phases: Phase 1 (strengthen guidance and tools such as threshold and impact assessment questionnaires), Phase 2 (transparency and engagement including Māori engagement guidance), and Phase 3 (oversight and governance, which may explore non‑binding audits, a public register of public service algorithms, and the creation of an oversight body). Other institutional stakeholders engaged with the Charter’s operation include the Public Service Commission, the Office of the Privacy Commissioner, and line agencies who become signatories; the Charter is positioned to complement existing laws and policies such as the Privacy Act and sector‑specific frameworks rather than replace them. Key public resources, templates and guidance are published in the data ethics section of the data toolkit.

Key Focus Areas

The Charter centralises several focus areas. Transparency requires plain‑English documentation of algorithmic decision‑making, publication of data and process information where lawful, and clear public explanation of how algorithms inform decisions. Partnership compels agencies to embed Te Ao Māori perspectives and honour Treaty principles in algorithm design and deployment, while noting that Māori data sovereignty may require separate, specialised consideration. People‑centric commitments require active engagement with affected communities, consulting those impacted by systems, and providing mechanisms to challenge decisions based on algorithms. Data commitments emphasise fitness for purpose, explicit identification and management of bias, and clear documentation of data provenance, collection, security and retention practices. Privacy, ethics and human rights commitments require regular peer review to detect unintended consequences, and remediation where issues are found. Human oversight insists on nomination of public contact points, channels for appeals or challenges, and transparency about the role and limits of human decision‑makers. Collectively these areas seek to balance public benefit from algorithmic analysis with protections for rights and the most vulnerable.

Implementation Framework

Implementation is risk‑based: agencies use an embedded risk matrix to determine whether the Charter should be applied (Low: could be applied; Moderate: should be applied; High: must be applied). Practical instruments provided include an Algorithm Threshold Assessment Questionnaire to screen algorithms, an Algorithm Impact Assessment (AIA) questionnaire and user guide for high‑risk systems, and report templates for documenting reviews. The implementation plan outlines capability building, the establishment of a community of practice, guidance for iwi and Māori engagement, and timelines for tools and guidance release. Agencies are expected to maintain records of assessments and peer reviews, nominate points of contact for public inquiries, and to publish explanatory material about algorithms where not prohibited by law. Over time the plan contemplates piloting non‑binding audits and developing oversight arrangements to increase assurance of Charter commitments.

Monitoring and Evaluation

The Charter envisages iterative review and evaluation. An independent one‑year review (completed in December 2021) evaluated uptake and recommended 24 practical considerations to improve guidance and implementation; Stats NZ and partner agencies committed to phased implementation of these recommendations. Monitoring activities in the implementation plan include community of practice activity metrics, publication of AIA outcomes (where lawful), capabilities built across signatories, and participation in Open Government Partnership milestones. Phase 3 contemplates broader assurance mechanisms such as voluntary audits or an oversight body; however, monitoring to date has primarily been through internal agency governance, peer review processes and public disclosure where permitted.

Penalties, Liability, and Appeals

The Charter itself is non‑statutory and contains no direct criminal or civil penalties. Liability for algorithmic harms remains governed by existing legal regimes (for example, the Privacy Act 2020, administrative law remedies, and statutes governing particular sectors). The Charter requires agencies to provide channels for challenge and appeal of algorithm‑informed decisions and to nominate a public contact for inquiries. Non‑compliance with the Charter may therefore result in reputational risk, internal administrative actions, public scrutiny, Official Information Act requests, and (in cases implicating privacy or other legal obligations) possible regulatory action under existing law. The implementation plan also contemplates exploring oversight mechanisms that could introduce new forms of assurance or encouragement to comply.

Relationship to Other Instruments

The Charter is explicitly framed to operate alongside and complement other instruments: it references the "Principles for the Safe and Effective Use of Data and Analytics" (Privacy Commissioner & Government Chief Data Steward, 2018), the Privacy Act 2020, the Official Information Act 1982, the Government Use of Artificial Intelligence report (2019), and sector frameworks such as the Privacy, Human Rights and Ethics Framework (Ministry of Social Development). It supports New Zealand’s Open Government Partnership commitments and seeks to align agency behaviours with established privacy, ethical and Treaty considerations. The Charter recognises some issues—most notably Māori data sovereignty—require dedicated, separate processes and instruments.

International Alignment

Although developed within New Zealand’s constitutional and Treaty context, the Charter aligns with international trends and soft law instruments emphasising transparency, human oversight, and risk‑based governance of automated decision systems. The Charter’s commitments are consistent with OECD AI principles and with comparable government initiatives that prioritize human‑centred design, explainability, and proportionality. New Zealand’s approach intentionally emphasises partnership with Indigenous communities, making explicit room for Te Ao Māori as a distinguishing feature while maintaining interoperability with international data governance norms.

Implementation Timeline

Sources and References