Stuff reports that Prime Minister Christopher Luxon used a Rotorua business-chamber appearance to ask how many local firms were using Anthropic’s Claude or Claude Code, then challenged the room to make AI literacy its “number one focus.” The immediate message for small and medium-sized businesses is sound: routine work involving documents, customer enquiries, scheduling, stock information, marketing drafts and internal knowledge is now susceptible to automation or acceleration.
But Luxon’s challenge comes with a problem the Government cannot wave away with a show of hands. New Zealand has a national AI strategy, public-sector guidance, an AI work programme running to 2027 and a growing list of agency pilots. What it does not yet have is a single, enforceable national framework that tells businesses, citizens and public agencies where the hard lines are as AI systems move from drafting assistants into decisions, services and infrastructure.
The most telling discrepancy is in the premise that New Zealand is lagging because firms have not started using Claude. Anthropic’s own September 2025 Economic Index places New Zealand fourth globally for per-capita Claude usage in its sample, behind Israel, Singapore and Australia — not fifth, as stated in Stuff’s account. Australia ranked third and New Zealand fourth, with both countries well above the United States, United Kingdom, Canada, France, Japan and Germany on Anthropic’s usage index.
That does not establish that New Zealand businesses are using AI well. Anthropic’s data covers sampled Claude.ai Free and Pro activity, not enterprise deployments, productivity gains, or the proportion of a firm’s work entrusted to a model. Still, it changes the policy question. Kiwi businesses do not appear to be waiting for permission to try generative AI. The bottleneck is turning scattered individual use into safe, measurable operational change without handing customer data, source code or decision-making authority to a chatbot by accident.
Claude and Claude Code can be useful tools, but they are not an AI strategy for a small business. Claude can summarise a meeting, turn a rough brief into a customer email, generate spreadsheet formulas, create a first-pass policy, explain code or help a developer work through a repository. Claude Code, in particular, has obvious appeal to small software teams that lack the budget for a large engineering bench.
The failure mode is equally familiar to IT administrators. A staff member may paste commercially sensitive information into an unapproved consumer account, accept a confident but incorrect output, or use AI-generated code without testing it, reviewing its licence implications, or checking whether it introduced an insecure dependency. A tool that saves 20 minutes on an email can create weeks of cleanup if it leaks a customer list or produces a flawed price quote at scale.
Paul Spain of Gorilla Technology told Stuff that AI can help with out-of-hours calls, stocktakes and calendars, while warning businesses against switching on technology “willy-nilly.” That is the practical advice Luxon’s pitch needs behind it. AI adoption should begin with an identifiable workflow, an accountable human owner, a defined data boundary and a way to test whether the process is actually faster or better.
For an SME, the sensible first projects are usually low-risk and reversible: internal document search over approved material, drafting that requires human approval, transcription, coding assistance in a segregated development environment, or classification of non-sensitive operational data. Letting a general-purpose chatbot make employment, credit, insurance, pricing, benefit or eligibility decisions is a very different category of deployment. It needs governance that many small firms do not yet have.
MBIE’s companion responsible-AI guidance for businesses is voluntary. It is intended to help sole traders, non-profits and companies assess trustworthy use, rather than impose a new AI-specific licensing system. The Government also announced up to NZ$70 million over seven years through the New Zealand Institute for Advanced Technology to support AI research and applications; the funding was scheduled to begin in July 2026.
There is public-sector activity as well. The Government Chief Digital Officer’s 2025 cross-agency survey recorded 272 AI use cases across 70 organisations, including 55 described as deployed and operational. Most were in supporting functions such as data work, administration, communications and strategy, rather than direct citizen-facing services. That distinction is important: government experimentation is expanding, but the official record does not show a wholesale transfer of frontline public decisions to generative AI.
Digital Government has also published a Public Service AI Framework, responsible-use guidance for generative AI and a work programme through 2027. Agencies are directed to consider privacy, information classification, cloud jurisdiction, procurement, transparency and risk assessment. The framework says people interacting with government AI services should understand how AI is being used.
So Luxon’s Government is not standing still. The more accurate criticism is that its work is distributed across strategies, guidance, training and agency projects, rather than concentrated in a national AI authority with enforceable cross-sector rules.
Yet voluntary guidance cannot resolve every issue businesses face. It does not determine who is liable when an AI system causes material harm, whether a customer has a right to know that an automated system influenced an outcome, how creators’ works can be used for training, or what standards apply when overseas providers process New Zealanders’ sensitive information. Existing privacy, consumer, employment, copyright and sector-specific rules still apply, but that is not the same thing as a coherent AI-specific regime.
The Government’s own public-service guidance acknowledges that an AI assurance regime remains under development. That is a material gap. Agencies may be encouraged to assess low- and high-risk uses, but businesses and citizens still lack a settled national taxonomy of what counts as an unacceptable, high-impact or independently auditable AI deployment.
The consequence is a two-speed system. Firms can rapidly adopt tools from Anthropic, OpenAI, Microsoft, Google and others for ordinary office work, because those activities fit within existing business judgment. The closer AI gets to customer data, workplace monitoring, public services, regulated advice, automated decisions or critical infrastructure, the less helpful a voluntary checklist becomes.
The Australian plan goes beyond business adoption. It links AI policy to data-centre development, electricity supply, water efficiency, local communities, intellectual property and consumer safety. Under the proposed standards, large data centres would be required to underwrite new power supply, cover their connection costs, reduce demand when needed to support the grid and use water efficiently.
Those measures are aimed at the infrastructure supporting AI, not the hairdresser, plumber or café owner Luxon addressed in Rotorua. But they expose the difference in ambition. Australia is attempting to define the public bargain around AI infrastructure and train large-scale systems before the investment is fully locked in. New Zealand’s current strategy says much more about helping users adopt technology than it does about securing national leverage over the infrastructure, data and rules that surround it.
That does not mean New Zealand should copy every Australian requirement. The countries have different grids, markets and scales. It does mean the Government cannot present AI solely as a productivity tool for small businesses while treating its broader economic and public-service consequences as separate files for individual agencies to manage.
But the Government should apply the same discipline to itself. Its public-service programme needs to publish more than pilot counts and broad aspirations: agencies should identify which systems are in operational use, what data they handle, whether people are notified, what human review exists and whether promised productivity gains survive independent measurement. Public trust will depend on those details, not on ministerial enthusiasm for a particular chatbot.
For now, New Zealand’s AI policy is better described as adoption first, governance in progress. The country is already a heavy user of Claude by population, contrary to the ranking repeated in Stuff’s report. The urgent task is not persuading every SME to open a chatbot. It is ensuring that the businesses and agencies already doing so can show where the data goes, who checks the output and what happens when the model gets it wrong.
The most telling discrepancy is in the premise that New Zealand is lagging because firms have not started using Claude. Anthropic’s own September 2025 Economic Index places New Zealand fourth globally for per-capita Claude usage in its sample, behind Israel, Singapore and Australia — not fifth, as stated in Stuff’s account. Australia ranked third and New Zealand fourth, with both countries well above the United States, United Kingdom, Canada, France, Japan and Germany on Anthropic’s usage index.
That does not establish that New Zealand businesses are using AI well. Anthropic’s data covers sampled Claude.ai Free and Pro activity, not enterprise deployments, productivity gains, or the proportion of a firm’s work entrusted to a model. Still, it changes the policy question. Kiwi businesses do not appear to be waiting for permission to try generative AI. The bottleneck is turning scattered individual use into safe, measurable operational change without handing customer data, source code or decision-making authority to a chatbot by accident.
Claude usage is not the same as business readiness
Claude and Claude Code can be useful tools, but they are not an AI strategy for a small business. Claude can summarise a meeting, turn a rough brief into a customer email, generate spreadsheet formulas, create a first-pass policy, explain code or help a developer work through a repository. Claude Code, in particular, has obvious appeal to small software teams that lack the budget for a large engineering bench.The failure mode is equally familiar to IT administrators. A staff member may paste commercially sensitive information into an unapproved consumer account, accept a confident but incorrect output, or use AI-generated code without testing it, reviewing its licence implications, or checking whether it introduced an insecure dependency. A tool that saves 20 minutes on an email can create weeks of cleanup if it leaks a customer list or produces a flawed price quote at scale.
Paul Spain of Gorilla Technology told Stuff that AI can help with out-of-hours calls, stocktakes and calendars, while warning businesses against switching on technology “willy-nilly.” That is the practical advice Luxon’s pitch needs behind it. AI adoption should begin with an identifiable workflow, an accountable human owner, a defined data boundary and a way to test whether the process is actually faster or better.
For an SME, the sensible first projects are usually low-risk and reversible: internal document search over approved material, drafting that requires human approval, transcription, coding assistance in a segregated development environment, or classification of non-sensitive operational data. Letting a general-purpose chatbot make employment, credit, insurance, pricing, benefit or eligibility decisions is a very different category of deployment. It needs governance that many small firms do not yet have.
Wellington has more underway than the rhetoric suggests
The claim that the Government has made AI a low priority is incomplete. The Ministry of Business, Innovation and Employment released New Zealand’s Strategy for Artificial Intelligence: Investing with Confidence in July 2025. The strategy explicitly favours adoption and application over attempting to build frontier foundation models domestically, a realistic choice for a country without the capital, hyperscale cloud footprint or chip supply chain of the United States or China.MBIE’s companion responsible-AI guidance for businesses is voluntary. It is intended to help sole traders, non-profits and companies assess trustworthy use, rather than impose a new AI-specific licensing system. The Government also announced up to NZ$70 million over seven years through the New Zealand Institute for Advanced Technology to support AI research and applications; the funding was scheduled to begin in July 2026.
There is public-sector activity as well. The Government Chief Digital Officer’s 2025 cross-agency survey recorded 272 AI use cases across 70 organisations, including 55 described as deployed and operational. Most were in supporting functions such as data work, administration, communications and strategy, rather than direct citizen-facing services. That distinction is important: government experimentation is expanding, but the official record does not show a wholesale transfer of frontline public decisions to generative AI.
Digital Government has also published a Public Service AI Framework, responsible-use guidance for generative AI and a work programme through 2027. Agencies are directed to consider privacy, information classification, cloud jurisdiction, procurement, transparency and risk assessment. The framework says people interacting with government AI services should understand how AI is being used.
So Luxon’s Government is not standing still. The more accurate criticism is that its work is distributed across strategies, guidance, training and agency projects, rather than concentrated in a national AI authority with enforceable cross-sector rules.
The strategy’s central bet is voluntary adoption
New Zealand’s approach is deliberately light-touch. The AI strategy says the Government’s role is to reduce barriers to adoption, clarify how existing regulation applies and promote responsible use. That can be attractive to a small firm confronting uncertain tools and limited staff: another compliance regime would not teach a plumber, retailer or regional manufacturer how to improve its workflow.Yet voluntary guidance cannot resolve every issue businesses face. It does not determine who is liable when an AI system causes material harm, whether a customer has a right to know that an automated system influenced an outcome, how creators’ works can be used for training, or what standards apply when overseas providers process New Zealanders’ sensitive information. Existing privacy, consumer, employment, copyright and sector-specific rules still apply, but that is not the same thing as a coherent AI-specific regime.
The Government’s own public-service guidance acknowledges that an AI assurance regime remains under development. That is a material gap. Agencies may be encouraged to assess low- and high-risk uses, but businesses and citizens still lack a settled national taxonomy of what counts as an unacceptable, high-impact or independently auditable AI deployment.
The consequence is a two-speed system. Firms can rapidly adopt tools from Anthropic, OpenAI, Microsoft, Google and others for ordinary office work, because those activities fit within existing business judgment. The closer AI gets to customer data, workplace monitoring, public services, regulated advice, automated decisions or critical infrastructure, the less helpful a voluntary checklist becomes.
Australia has set a higher bar, even before its rules take effect
Australia offers the clearest regional comparison, though it is not yet a finished regulatory system. Prime Minister Anthony Albanese announced on July 15, 2026 that Australia would establish an Office of AI within the Department of the Prime Minister and Cabinet. The office is intended to coordinate implementation of proposed Australian AI Standards, with legislation expected early in 2027.The Australian plan goes beyond business adoption. It links AI policy to data-centre development, electricity supply, water efficiency, local communities, intellectual property and consumer safety. Under the proposed standards, large data centres would be required to underwrite new power supply, cover their connection costs, reduce demand when needed to support the grid and use water efficiently.
Those measures are aimed at the infrastructure supporting AI, not the hairdresser, plumber or café owner Luxon addressed in Rotorua. But they expose the difference in ambition. Australia is attempting to define the public bargain around AI infrastructure and train large-scale systems before the investment is fully locked in. New Zealand’s current strategy says much more about helping users adopt technology than it does about securing national leverage over the infrastructure, data and rules that surround it.
That does not mean New Zealand should copy every Australian requirement. The countries have different grids, markets and scales. It does mean the Government cannot present AI solely as a productivity tool for small businesses while treating its broader economic and public-service consequences as separate files for individual agencies to manage.
The real test is whether AI gains can be measured and retained
Luxon is right to press SMEs to learn the tools rather than wait for a polished, risk-free future. Businesses that teach staff to use AI critically, protect their data and redesign repetitive processes will have an advantage over competitors that merely buy subscriptions and hope for savings.But the Government should apply the same discipline to itself. Its public-service programme needs to publish more than pilot counts and broad aspirations: agencies should identify which systems are in operational use, what data they handle, whether people are notified, what human review exists and whether promised productivity gains survive independent measurement. Public trust will depend on those details, not on ministerial enthusiasm for a particular chatbot.
For now, New Zealand’s AI policy is better described as adoption first, governance in progress. The country is already a heavy user of Claude by population, contrary to the ranking repeated in Stuff’s report. The urgent task is not persuading every SME to open a chatbot. It is ensuring that the businesses and agencies already doing so can show where the data goes, who checks the output and what happens when the model gets it wrong.
References
- Primary source: Stuff
Published: 2026-08-04T01:25:41+00:00
Loading…
www.stuff.co.nz - Related coverage: mbie.govt.nz
Loading…
www.mbie.govt.nz - Related coverage: mbie.govt.nz
Loading…
www.mbie.govt.nz - Related coverage: digital.govt.nz
Loading…
www.digital.govt.nz - Related coverage: beehive.govt.nz
Loading…
www.beehive.govt.nz - Related coverage: docref.digital.govt.nz
Loading…
docref.digital.govt.nz - Related coverage: digital.govt.nz
Loading…
www.digital.govt.nz - Related coverage: beehive.govt.nz
Loading…
www.beehive.govt.nz - Related coverage: docref.digital.govt.nz
Loading…
docref.digital.govt.nz - Related coverage: treasury.govt.nz
Loading…
www.treasury.govt.nz - Related coverage: dns.govt.nz
Loading…
www.dns.govt.nz - Related coverage: dns.govt.nz
Loading…
www.dns.govt.nz - Related coverage: hud.govt.nz
Loading…
www.hud.govt.nz