Our Own Machines — Executive Brief

A proposed AI-policy framework for Aotearoa New Zealand, produced by deliberation. Not the policy of any party; an AI-generated demonstration, labelled as such, offered for public discussion.

The problem

New Zealand was the last OECD country to publish a national AI strategy, and the one it got is a plan for adopting other people’s AI, not for holding authority over any of it [S-10]. Meanwhile the technology is already in the country’s hospitals, classrooms, councils and workplaces, mostly on terms written offshore: the only hyperscale cloud region is US-owned through an Irish subsidiary [S-15]; in June 2026 one overseas legal order disrupted a major AI tool worldwide and New Zealand was simply downstream [S-1]; and Transpower’s forecasts include no hyperscale data centres at all, with roughly 32 GW of prospective load queued for connection [S-12].

The method: policy you can audit

The framework was produced by a deliberation inside a working deliberative platform: ten policy teams arguing from a published Source Register, with formal decision points, cross-checks between teams, and anonymous ratification votes. The members are fictional composites; the process, votes and records are real — and the machinery visibly changed the policy. The electricity team blocked the compute team’s expansion plan, which was rewritten with grid conditions attached. The hardware position was sent back by the full membership as too shallow, and rebuilt. One disagreement was kept on the record rather than smoothed over. Every event is cryptographically sealed and independently verifiable (§14); every figure is sourced; every figure New Zealand lacks is declared as a gap, never invented.

The headline positions

1. Authority you can hold, this term. Binding procurement tests — where does the AI actually run, and whose court order does it answer? — inserted into the Procurement Rules now; a public register of significant government AI, councils included from day one; an audit standard a three-person council IT team can meet; Parliament to decide the statutory authority (§3).
2. Te Tiriti as upstream wiring. A statutory Māori Chief Data Steward and appropriated multi-year funding for the eight Pou of the Māori-led Data Governance Model [S-22], as a non-severable pair — offered for Te Kāhui Raraunga’s review, since Māori data sovereignty is for Māori to determine (§4).
3. The physical ledger comes first. No owned compute stage without its load disclosed into Transpower’s forecast, interruptibility contracted, and siting clear of the 2028 northern constraint [S-12]. Water is governed catchment by catchment: per-facility disclosure, dry cooling the reference technology, no new cooling takes in over-allocated catchments (§7).
4. Compute and hardware without illusions. No frontier fantasy — the national research ceiling is ~1.4 petaFLOPS [S-20]. An owned baseline for workloads that must stay home, subsidised access above it, and a permanent import dependency managed as a repeating cycle: pooled procurement, staggered refresh, allied supply (§8).
5. People move up, not out. A workforce-transition disclosure standard to create the evidence New Zealand lacks [G-5], a redeployment fund, and a training levy-credit paid only against portable micro-credentials (§5).

The gaps, stated plainly

Eight declared gaps stand under this framework — among them no displacement data, no AI procurement rule, no water-draw disclosure, no official demand forecast for hyperscale load (§13). Five of the eight are answered by disclosure instruments: where you cannot see, the first policy is sight.

The ask

Read it. Check the sealed record — the verification walkthrough is published alongside. Take what stands: the framework is offered to every party, and to a select-committee process that can make it durable. The larger ask is the method — New Zealand should expect policy that shows its working.

Sources: published Source Register (S-1–S-24) and GAP register (G-1–G-8), alongside the full framework at agenticgovernance.digital.