Banning super-AI is easy when nobody has to define it

Labour has found its AI soundbite. In a Stuff interview ahead of Election 2026, the party called the coalition’s AI readiness “terrifying” and repeated its call for a ban on hypothetical “super-AI” systems.

The criticism of the government has some merit. The ban does not. Nobody, including Labour, has said what super-AI is, how a ban would be enforced against systems hosted offshore, or what it would cost. That is a slogan, not a law. It also draws attention away from the AI problems businesses actually face.

A ban with no definition attached

When Labour launched its AI platform in September, it was reasonably grounded. Leader Chris Hipkins and spokesperson Reuben Davidson proposed an Office of AI inside DPMC, a wider remit for an online safety regulator, copyright rules so creators are paid when their work trains models, a deepfake porn ban, renewable energy requirements for data centres, and SME adoption support. Asked what it would cost, Davidson declined to give estimates and called it an “action plan”.

The super-AI ban has made that costing gap bigger. Superintelligence and artificial general intelligence are contested ideas even among researchers. A law has to say which systems fall inside the line and which fall outside it. Without that, firms cannot tell whether the next version of a tool they already pay for counts. New Zealand also cannot stop a frontier model trained in California from being used here, any more than it can ban the weather.

Davidson is right about the stakes. He has called AI “the most significant industrial revolution of our time” and said the government needs to map which jobs it will disrupt. That is a serious point. A ban on science fiction is not.

The real gap is in the back office

The live issues are far less dramatic. MBIE’s 2025 AI strategy recorded that 67% of larger firms used AI in 2024, while 68% of SMEs had no plans even to evaluate it, compared with 38% in Australia. It found 43% of non-users blamed a lack of expertise. That is a skills and confidence problem. A frontier ban does nothing about it.

The tools already inside accounting software, CRM platforms and HR systems raise ordinary questions. Who owns outputs trained on copyrighted material? What happens when customer data goes into an offshore model? Who is accountable when an automated decision goes wrong? In 2025, the Privacy Commissioner told officials four in five New Zealanders did not think current AI regulation was adequate, and suggested updating the Privacy Act as an achievable first step. That is a fix that could be written and passed. A super-AI ban cannot.

The coalition isn’t off the hook

The government should not feel smug. Its 2025 Cabinet paper said “business is best placed to decide how to incorporate AI”, and Technology Minister Judith Collins has kept intervention for acute risks or unlocking innovation. That is a defensible starting point. But New Zealand now ranks 32nd of 45 high-income countries on the Global Index on Responsible AI, behind Australia at 22nd. Light-touch should not mean leaving businesses to guess.

The state is more disciplined with its own AI use than it asks the private sector to be. The Stats NZ transparency register lists 13 approved government uses and three time-limited trials, each checked for safeguards and human oversight. Agencies get a checklist. Businesses get voluntary guidance.

University of Otago researchers have called for “a bipartisan effort to produce risk management-based AI regulation” and a national oversight body. They note that Labour’s platform is the most detailed of any party, and that most others had barely published anything. Labour deserves credit for turning up. It has then spent its attention on the weakest part of its own platform.

What business should watch

For larger firms already using AI, the copyright framework and data standards matter most. For data centre developers and big energy users, Labour’s rule that facilities secure their own renewable supply and reduce peak demand is concrete and could be costed. It deserves more scrutiny than any frontier ban. For SMEs, the question is whether either side will fund the advisory support that gets them started.

The election offers two choices. One is a business-led model that risks drifting. The other is a centralised Office of AI that nobody has costed. Neither side has priced its plan. If Labour wants to be taken seriously on AI, it should drop the super-AI ban and publish a costed Privacy Act update instead. If the coalition wants to keep its light-touch approach, it needs to show that light-touch still means firms know the rules.

Sources

• Stuff: Labour blasts government’s ‘terrifying’ lack of AI readiness, wants super-AI ban (2026-10-22)
• RNZ: Labour promises to set up AI regulator, copyright rules if elected (2026-09-09)
• RNZ: ‘The most significant industrial revolution of our time’: Labour’s Reuben Davidson on NZ’s AI future (2026-09-28)
• RNZ: Taking a light touch with AI’s ‘dangers and evils’ (2026-10-12)
• Stats NZ: Artificial Intelligence (AI) transparency register (2026-08-01)