Seymour’s AI worries skip the productivity question business needs answered

David Seymour’s anxieties about artificial intelligence made headlines again this week, with Stuff reporting on the three things that scare the ACT leader. Fair enough. Politicians should think about risk. But the more revealing part of the story is what is missing. Six weeks out from polling day, New Zealand is arguing about what could go wrong with AI and has no serious plan for making it go right.

For business owners, that gap matters more than any single fear. The firms that win from AI will not wait for Parliament to finish drawing guardrails.

Seymour is more pro-adoption than his fears suggest

To be fair to Seymour, he is not a technophobe. Last month he told RNZ he can’t see why AI shouldn’t write Cabinet papers, so long as there is a human in the loop. His office has run dummy exercises using AI to speed up policy drafting, and he wants the public sector to adopt it aggressively.

His worries are specific. He flagged the risk of AI empowering “lone wolves”, citing a Sydney man who used AI to help develop a cancer therapy for his dog, and has floated the possibility of self-reproducing intelligence by 2035. On regulation, he wants New Zealand to be a “fast follower” rather than a rule-setter.

That instinct is broadly right. A small economy writing bespoke AI law ahead of the EU, US and Australia would mostly create compliance costs for local firms. But “fast follower” is a regulatory stance, not a growth strategy. RNZ’s John Hartevelt put his finger on it: most party rhetoric so far focuses on managing risk rather than explaining how New Zealand captures the benefits, while other countries treat AI as strategic infrastructure.

The Government has already bet on AI without testing the bet

Here is the awkward bit. The Government has effectively priced AI productivity into its own budget. In May, Seymour’s AI guidance for regulators landed alongside the announcement of 8,700 public service job cuts, with AI expected to help fill the capability gaps. Experts, including Victoria University of Wellington’s Ali Knott, warned that generative AI in government needs rigorous evaluation by staff who actually understand the systems.

Treasury’s own work is sobering. In 2025 it said it had found no evidence of aggregate productivity growth from AI, despite positive effects inside individual firms. Back in 2024 it flagged New Zealand’s historically weak technology diffusion and low investment in intangible capital as specific barriers. Translation: the gains are real at the firm level, but this country is historically bad at spreading them.

Cutting headcount first and hoping AI fills the hole is not a productivity plan. It is a cost-saving exercise with a technology label attached.

Awareness is not the problem, conversion is

The private sector does not need persuading that AI exists. MBIE’s 2025 survey of 500 SMEs found 94% were aware of at least one AI tool, but adoption and confidence varied sharply. Firms said they wanted practical guidance and worked examples, not abstract rules. MBIE’s accompanying guidance promised a “proportionate risk-based approach”, sensibly avoiding the EU’s heavier regime.

Earlier evidence was encouraging. A 2024 Callaghan Innovation and AI Forum survey of 232 organisations found 67% were already using AI, 96% said it made workers more efficient, and only 8% reported it replacing staff. By late 2025 the mood had cooled. Greg Davidson, then Datacom chief executive, acknowledged many firms still weren’t seeing the promised productivity boost, and the same reporting warned the real disruption could come from small AI-native competitors running far leaner than incumbents.

That last point should worry boards more than any lone-wolf scenario. If a handful of people with good tools can match a 100-person operation, the threat to your margins is not rogue AI. It is a rival who worked out the workflow before you did.

What a real plan would contain

None of this is new. As far back as 2023, BusinessNZ urged cross-party work on frameworks that create safe pathways for productivity, growth and jobs. Three years on, the parties have at least found common ground on machinery, with Seymour endorsing Labour’s proposed Office of AI in principle. That is useful, but an office is not a plan.

A serious plan would measure adoption and output at the firm level, fund practical diffusion through existing channels, fix the data and privacy settings firms cite as blockers, and publish honest evaluations of the public sector’s own AI experiments before banking savings from them.

Whoever forms the next government should be judged on whether AI shows up in the productivity statistics, not on how many risks they can name. In the meantime, business owners should take the hint. The policy will arrive late. The competitors using AI well are already here.

Sources

• Stuff: Three things scare David Seymour about AI (2026-11-02)
• RNZ: ACT leader David Seymour doesn’t see why AI shouldn’t write government Cabinet papers (2026-10-02)
• RNZ: AI has finally entered the election campaign arena (2026-10-03)
• BusinessDesk: Experts warn over ‘light-touch’ approach in Seymour’s new AI plan (2026-05-28)
• Treasury: Official Information Act Response 20250500 – Future of AI (2025-08-13)
• Callaghan Innovation: AI in Action – September 2024 Report (2024-09)
• BusinessDesk: CEOs discover AI not a productivity magic bullet (2025-10-22)
• BusinessNZ: What businesses want from the incoming government (2023-10)