Year 3–5: Building Permanent Infrastructure (2028–2030)

Context

By 2028, New Zealand will have a national AI impact baseline, pilot evaluation data, and two years of experience operating the Transition Support Fund. This phase uses that evidence to scale what works and build the permanent institutions that will outlast any single government. The defining challenge is political continuity: the deepest structural reforms take longer than an electoral cycle, and institutions are more durable than policies.

Establishing the NZ AI Commission

The most important institutional act of this phase is establishing a standing NZ AI Commission, modelled explicitly on the Climate Change Commission but adapted for AI's faster pace of change. The Climate Change Commission has demonstrated that an independent statutory body, with a clear mandate, transparent methodology, and the authority to issue public recommendations, can shift the terms of political debate and maintain policy direction across governments of different composition.

The AI Commission's core functions should include: annual state-of-AI-in-NZ reports drawing on updated impact assessment data; independent evaluation of government AI programmes; public reporting on AI incidents and systemic risks; and five-yearly transition pathway recommendations to Parliament. Unlike a ministry, the Commission should have no operational AI responsibilities — its job is to watch, measure, and advise, not to implement. This separation preserves independence.

Scaling Education Reform

The Year 1–2 literacy pilots will have produced outcome data across decile range and curriculum type. Year 3 is the decision point: which elements get embedded into the national curriculum, and what support does the teaching workforce need to deliver them at scale?

Curriculum reform at this scale requires a minimum three-year lead time from design to full implementation. The Year 3 decision should trigger an immediate curriculum development process, with phased rollout targeting full implementation by 2031. Parallel investment in initial teacher education (ITE) and continuing professional development (CPD) is non-negotiable — curriculum change without teacher capability change produces compliance without learning.

Tertiary education reform should run in parallel: universal digital skills pathways across all NZQA qualifications, not as standalone add-ons but integrated into existing credentials. Funding model changes may be needed to make short-course reskilling economically viable for both providers and learners.

Sovereign Compute Infrastructure

If the Year 1–2 sovereign inference pilot demonstrates viable performance and cost metrics, Year 3 is the right moment to commission dedicated infrastructure. An investment in the range of $200–500M over five years — potentially structured as a public-private partnership with NeSI, REANNZ, and commercial data centre operators — would give NZ meaningful capability for sensitive government AI workloads and a platform for public-interest AI research.

The infrastructure decision should be evidence-driven, not ideological. If the pilot data shows that commercial APIs offer equivalent data sovereignty guarantees at lower cost, that is a valid finding. The goal is not sovereignty for its own sake but appropriate control over sensitive workloads and resilience against supplier dependency.

Data Sovereignty Legislation

The Regulatory Audit from Year 1 should have identified the legislative gaps; Year 3 is the window to close the most significant ones. Priority legislation should include: an AI Systems Register requiring transparency about automated decision-making in government and regulated industries; amendments to the Privacy Act addressing algorithmic profiling and synthetic data; and a Data Sovereignty Act establishing NZ's framework for controlling cross-border data flows, with particular attention to te ao Māori data governance principles (drawing on the Māori Data Sovereignty Network's existing framework).

Sector-Specific Transition Plans

The National AI Impact Assessment will have identified the highest-exposure sectors. By Year 3, each should have a government-supported transition plan, developed in genuine partnership with unions, industry bodies, and affected workers. These plans should be specific — not aspirational statements about "upskilling" but concrete pathway maps: what roles are declining, what roles are growing, what the training pathway between them looks like, and what income support is available during transition.

The first tranche of plans should cover the five highest-exposure sectors identified in the Year 1 assessment. A further tranche should follow by Year 5 covering the next tier of exposure.

First International AI Cooperation Agreements

NZ cannot govern AI in isolation. Year 3–5 should see the first substantive international AI cooperation agreements, prioritising: mutual recognition of AI safety standards with Australia, the EU AI Act framework, and Pacific regional partners; data sharing agreements for AI research with Five Eyes partners (with appropriate sovereignty carve-outs); and engagement with the UN's AI governance processes. The goal is to shape international norms from a position of credible domestic policy, not to import frameworks designed for larger economies.