The Green Party released a policy document on Monday afternoon titled "Safe, sovereign AI," with co-leader Chlöe Swarbrick contending that governments have a responsibility to take an active role in shaping how artificial intelligence develops — spreading its benefits widely while guarding against the harms it could bring.
The party's stance on large-scale AI data centres has shifted from a fixed one-year building moratorium to an indefinite halt, with construction to remain suspended until a regulatory framework is established. The proposal that emerged from the planning process differs from the July position in one respect: where an outright ban on new hyperscale data centres was initially sought, the document now frames the measure as a consent pause, to remain in place only until a regulatory framework is established.
The party's official spokesperson indicated that the policy had shifted, with the updated stance permitting large-scale centres to receive consent on a faster timeline than the original one-year moratorium had envisaged. The timeline carries no defined endpoint, leaving open the possibility of a later conclusion.
Where Labour has put forward an AI plan backed by a dedicated Office of AI, National's current position rests on the coalition's principles-based framework, drawing on legislation and regulations already in place.
Swarbrick put forward the position that her party would bring oversight and public-interest principles to bear on the matter. The position put forward holds that artificial intelligence ought to serve the interests of people and the natural world rather than the reverse.
The Green Party's proposed AI framework would keep human oversight and democratic decision-making intact, with local communities holding authority over their own data, technological operations confined within environmental limits, economic benefits spread across the broader population rather than concentrated among a select few, and all AI systems subject to transparency requirements and accountability mechanisms.
Proposed legislation from the Greens would oblige operators of hyperscale AI data centres to undergo environmental impact assessments before construction, while also requiring each new facility to bring sufficient additional renewable generation capacity online to cover its entire electricity draw. Four provisions would take effect under the plan: caps on water consumption, obligations to report usage publicly, processes for consulting affected communities, and a requirement to obtain local consent.
The document's provisions on artificial intelligence specify that automated systems alone are not to be permitted to reach binding conclusions in any public-service context where the outcome could carry real weight for the individuals affected. The document outlined that those making AI-related decisions faced accountability both at the time and going forward.
Under the party's proposals, government agencies would face mandatory disclosure requirements covering which AI systems they deploy, alongside obligations to make available detailed information about how and where artificial intelligence has informed policy development and decision-making.
The plan incorporates both a provision allowing individuals to withdraw their participation and a firm restriction on sharing data with outside parties. The policy states that anyone subject to automated decision-making retains the right to contest those outcomes and to remove themselves from such processes, and that neither personal information held in trust by the government nor classified government data may be passed to third parties, whether for the purpose of training models or running inference.
The proposal calls for public data to be stored domestically, and outlines the creation of an AI office housed initially within the Department of Prime Minister and Cabinet, with a pathway toward establishing it as a standalone AI Commission carrying powers to conduct assurance work, audits, and public education.
The policy document outlines requirements around consent for AI training data, compensation arrangements, and financial penalties linked to intellectual property drawn upon in building AI systems, together with Māori data sovereignty provisions that would require co-governance and co-development of Māori data and intellectual property. Swarbrick contended that neither the dire scenarios circulated by technology billionaires nor the financial-bubble collapse projections put forward by economists represent inevitable outcomes, that the role of government is precisely to regulate, distribute benefits on equitable terms, and constrain harms, and that the Green Party would work to protect the natural environment, people in creative industries, data sovereignty, the wider workforce, and national interests from those pursuing short-term extraction at the public's expense.
