AI Governance for NZ SMBs: Writing an AI Policy and Why ISO 42001 Matters

Most New Zealand businesses now use AI in some form, yet far fewer have written down the rules for using it. This guide explains why an AI policy matters for a NZ business, what a good policy contains, and where ISO 42001, the international standard for AI management, fits for smaller organisations.

It is written for owners and general managers rather than lawyers or data scientists. OxygenIT has run IT for New Zealand businesses from Christchurch since 2005. We hold ISO 42001 certification as at 12 May 2026 for AI management, audited by Compass Assurance Services, alongside ISO 27001 for information security, and we write AI policies for clients while running our own AI use under the same standard.

Does my NZ business need an AI policy?

This guide focuses on the policy side. If you want the bigger picture first, read our complete guide to what AI governance is and the six pillars behind it.

Yes. Most NZ organisations already use AI, but far fewer have an AI policy, and fewer again have formal ethics or safety guidelines. Staff are using these tools today, with or without rules, so the real choice is whether that use is governed.

The same index reports that 52 percent of NZ organisations cite shadow AI as a concern, meaning staff using AI tools the business has not approved and cannot see. The concern is well founded. When a staff member pastes a client contract into a personal chatbot account to summarise it, your business has just disclosed confidential information to a third party under terms nobody has read.

The pattern behind those numbers is consistent: adoption ran ahead of governance. An account manager drafting proposals with ChatGPT, a coordinator running an AI notetaker in client meetings and a finance lead summarising spreadsheets with Copilot are all routine in 2026. If Microsoft Copilot is on your roadmap, our Microsoft Copilot readiness guide covers the permission and data hygiene work to complete before rollout. The question is no longer whether your team uses AI. It is whether anyone has written down the rules.

What should an AI policy include?

A useful AI policy fits on a few pages and answers practical questions: which tools are approved, what data may go into them, who reviews outputs before they leave the business, and what happens when AI gets something wrong. If a document does not cover the eight areas below, it is not finished.

AI policy checklist

  • Acceptable use and approved tools. Name the AI tools staff may use for work, on which accounts, and for which tasks. Everything else needs approval before first use.
  • Data input rules. The single most important clause: no personal information and no client confidential data in unapproved tools, ever. Spell out what counts as each, with examples from your own workflows.
  • Privacy obligations. Tie the policy to the Privacy Act 2020 so staff understand that pasting customer details into a public chatbot is a disclosure, not a shortcut.
  • Human review. Any AI output that leaves the business, including proposals, reports, advice and client emails, is checked by a person who is accountable for it.
  • Disclosure rules. State when you tell clients that AI was involved in producing work, and when a contract or plain good faith requires it.
  • Incident path. Who staff tell when AI goes wrong, such as leaked data, a hallucinated fact sent to a client or a biased output, and how quickly.
  • Training expectations. The induction and refresher programme every staff member completes, so the policy becomes a skill rather than a document.
  • Review cadence and owner. A named policy owner and a review date at least annually, because the tools change monthly.

A free AI policy template can seed this list, but it only becomes useful once the named tools, the data classes and the owner are yours. Keep the language plain, prioritise the rules that protect client data, and write for reading rather than filing. That is the difference between governance and paperwork.

What is ISO 42001?

ISO 42001 is the first international, certifiable standard for managing artificial intelligence, published by ISO. It requires an AI management system: a documented, auditable framework governing how your organisation develops or uses AI. Certification is granted only after an external audit, the same model long used for ISO 27001 in information security.

If you want ISO 42001 explained in one line: it does for AI what ISO 27001 did for information security. Rather than prescribing particular tools, it defines a management system with four working parts:

  • Risk assessment per AI use case. Every place AI is used, from marketing copy to client reporting, is identified and assessed for its impact on customers, staff and the business.
  • Defined accountabilities. Named people own each AI decision, so when an output causes harm there is no doubt about who answers for it.
  • Lifecycle controls. Governance follows each AI system from selection and approval through operation, monitoring and retirement, not just at purchase.
  • Continual improvement. The whole system is measured, reviewed and corrected on a cycle, because models, tools and risks keep moving.

Certification is confirmed through an external audit by an accredited certification body, with surveillance audits after that, so an ISO 42001 certificate is independently tested evidence rather than a self assessment. OxygenIT went through that process with Compass Assurance Services and holds the certificate, which remains rare for an IT provider in New Zealand, so this guide describes a system we operate, not one we have only read about.

Why does ISO 42001 matter for an SMB using AI?

For most SMBs the value of ISO 42001 is the discipline rather than the certificate. The standard forces a business to list its AI use cases, assess the risk of each and assign owners, which is exactly the structure that shuts down shadow AI. The certificate itself matters when clients, insurers and procurement teams start asking.

Client and public trust is the first reason. Your customers are already asking how businesses protect their data, and questions about AI use are following quickly. Insurers and procurement teams that spent the last five years asking about MFA and backups are starting to ask about AI governance in the same questionnaires. A business that can point to a recognised structure answers in one line instead of twenty.

The second reason is internal. The working parts of the standard, a register of use cases, a risk assessment for each and named owners, are precisely what stops shadow AI from taking root. Staff get a clear list of approved tools and a route for requesting new ones, so the incentive to go around IT disappears.

Should an SMB certify? Usually not, at least not first. Certification takes budget and management attention that a twenty person firm rarely needs to spend, because the leverage sits in the structure rather than the badge. The practical model for AI governance in an SMB is to borrow the framework and work with certified partners for the heavy lifting. That is how we run our AI implementation services, and we cover the small business angle in more depth in our guide to AI governance for small business under ISO 42001.

How do the NZ Privacy Act and MBIE Responsible AI Guidance apply?

New Zealand has no AI specific statute yet, but existing law fully applies. The Privacy Act 2020 covers any personal information that goes into a prompt or comes out of a model, the Office of the Privacy Commissioner has published expectations for AI use involving personal information, and MBIE has published voluntary Responsible AI Guidance for businesses.

Start with the Privacy Act, because it carries legal weight today. Its information privacy principles apply to personal information wherever it travels, including into an AI prompt. Typing a customer name, complaint history or health detail into a public AI tool is a use and disclosure of personal information, and the party accountable is your business, not the AI vendor. Outputs count too: when a model generates information about an identifiable person, you remain responsible for its accuracy and how it is used. The Office of the Privacy Commissioner expects organisations to think before adopting AI, including assessing privacy impacts and keeping human review in the loop where personal information is involved.

MBIE has published Responsible AI Guidance for businesses, setting out how New Zealand firms can adopt AI safely within existing law. The guidance is voluntary, and that is not a reason to ignore it. It is a clear signal of the direction responsible AI in NZ is heading, and it tells you what regulators and large customers will treat as good practice long before any statute lands.

For an owner the takeaway is short: your AI policy is where these obligations become instructions. The Privacy Act supplies the legal floor, the MBIE guidance supplies the direction of travel, and the policy translates both into what your staff may and may not do on a Tuesday afternoon.

AI policy NZ business: frequently asked questions

Is an AI policy legally required in New Zealand?

No statute forces you to have one. There is no AI specific law in New Zealand yet. The Privacy Act 2020 still applies to personal information used with AI, and directors have general duties to manage risk. A written policy is the practical way to show you take both seriously.

What is shadow AI?

Shadow AI is staff using AI tools the business has not approved or does not know about, such as personal ChatGPT accounts used for work tasks. It matters because company or client data can end up in tools with no contract, no controls and no audit trail. It is one of the AI risks NZ organisations name most often.

Can we just ban AI instead of writing a policy?

A ban rarely works. Staff see the productivity gain and keep using personal accounts on personal devices, which pushes usage underground and makes the data risk worse. A policy with approved tools and clear input rules gives people a safe route, so the business gets the benefit while keeping control of its information.

Is a free AI policy template enough?

A template is a starting point, not a finished policy. Generic documents miss the tools your team actually uses, your client confidentiality obligations and your industry rules. Fit the template to your real workflows, name an owner, train staff on it and set a review date. A policy nobody follows offers no protection.

Does ISO 42001 apply to small firms?

Yes. The standard scales to organisation size, and certification is realistic for a small firm with the right support, though most SMBs do not need to certify. The practical route is to borrow the structure, meaning a use case register, risk assessment and named accountabilities, and to work with partners that hold the certificate.

How does OxygenIT help with AI governance?

We write AI policies and governance frameworks for NZ businesses, run staff training and set up the technical controls that make a policy stick. OxygenIT is certified as at 12 May 2026 to ISO 42001 and ISO 27001, so the advice comes from a system we operate ourselves every day. Contact us to talk through where to start.

Govern AI safely with OxygenIT (ISO 42001 certified)

OxygenIT is certified to ISO 42001 for AI management, audited by Compass Assurance Services, alongside ISO 27001 for information security, as at 12 May 2026. We have operated from Christchurch since 2005, with coverage in Wellington, and we write AI policies and governance frameworks for clients while running our own AI use under the same standard. You can read more about our certifications on our about page.

Book a discovery call and we will map where AI is already in use across your business, draft the policy with you and put the governance behind it. Or call us on 0800 101 095.