The short version
- Staff are using AI whether or not you have a policy. A policy makes it safe and useful instead of secret and risky.
- One page: approved tools, what never goes in, how outputs are checked, who owns it, how it is reviewed.
- ISO/IEC 42001 is the standard behind it — use its structure, certify only if a customer or regulator asks.
- The technical controls — tenant-bound tools, permissions, logging — are what make the page true.
Every business we visit has staff using AI. About one in five has told them how. The rest have a policy by default: whatever each person decided, with whatever data they pasted. This is the page that replaces that, and it takes an hour to adopt.
The one-page AI policy
1. Approved tools
Name them. Typically: Microsoft 365 Copilot (inside your tenant), the AI features inside systems you already license (Xero, your practice or job system), and — for general, non-confidential tasks only — one named consumer tool with a business account. Anything else needs approval.
2. What never goes into a consumer tool
Customer and staff personal information, financial details, legal matters, health information, passwords, unreleased commercial information. If it would be a Privacy Act problem in an email to a stranger, it is a problem in a chat box.
3. Outputs are drafts
Everything AI produces is checked by a person who is responsible for it before it goes to a customer, a regulator or a system of record. Citations and figures are verified against the source.
4. Transparency
Where AI materially contributes to advice, a document or a decision about a person, we say so — in terms of engagement and, where appropriate, in the work itself.
5. Ownership and review
A named owner. A review every six months, or when a new tool is adopted. A simple log of which tools are in use and for what.
6. If something goes wrong
Data pasted where it should not have been, an AI-drafted error reaching a customer: tell the owner the same day. Mistakes reported fast are fixable; mistakes hidden are breaches.
What ISO/IEC 42001 adds
ISO 42001 is the management-system standard for AI. Stripped of the standards language, it asks an organisation to:
| ISO 42001 asks |
In plain English |
On the one page |
| Context and leadership |
Why you use AI and who is accountable |
Section 5 |
| Risk and impact assessment |
What could go wrong for whom, before deployment |
Sections 2 and 3, per tool |
| Controls |
Technical and procedural safeguards |
Sections 1–3 plus your IT controls |
| Transparency and human oversight |
People know, and a person remains responsible |
Sections 3 and 4 |
| Monitoring and improvement |
Review, log, learn |
Sections 5 and 6 |
A 20-person business does not need certification. It needs the structure, so that when a large customer’s procurement form asks “describe your AI governance”, the answer is a page rather than a pause. Be AI aligns every engagement to ISO 42001 and the NIST AI Risk Management Framework for exactly that reason.
The controls that make the page true
- Tenant-bound tools for anything with your data — Copilot, not a free chat window.
- Permissions reviewed before AI is switched on, because AI surfaces what people can already open.
- Sensitivity labels on confidential folders, and data-loss policies that stop labelled content leaving.
- Logging of AI use where the tool provides it, kept with the IT partner.
Questions people ask
Can staff use ChatGPT at work?
With a policy, yes for general tasks u2014 drafting a generic email, explaining a concept, brainstorming. Without customer, staff, financial, legal or health information, and never as the system of record. For anything involving your data, use a tool inside your Microsoft tenant such as Copilot.
What is ISO/IEC 42001?
The international standard for an AI management system, published in 2023. It is to AI what ISO 27001 is to information security: a structured way to govern, risk-assess, document and improve how an organisation uses AI. Certification is for larger organisations; the structure is useful at any size.
Do we need to tell customers we use AI?
If AI materially shapes something a customer relies on u2014 advice, a document, a decision about them u2014 transparency is expected and, in regulated professions, often required. A line in your terms of engagement is the usual approach.
Who should own the AI policy?
A director or the general manager, with the IT partner responsible for the technical controls. It should not sit with the most enthusiastic user.