A complete policy, below, free and ungated. Fill in a name, a date and the tools you actually approve, and it is ready to use.
An AI use policy sets out which AI tools your team may use for work, what information must never be put into them, and who checks what comes back out before anyone relies on it. It should fit on one page, name a person who owns it, and describe behaviour rather than products, because the products change faster than the document will.
If anyone in your business already uses AI for work, you already have a policy. It is unwritten, it is different in each person's head, and you cannot see it. Writing it down is how you find out that two people had opposite ideas about what was acceptable.
Replace the bracketed name, date and tools list with your own. Twelve short clauses, written to be adopted rather than admired.
Everyone who works here, on any AI tool, for any work task, whether it is one we pay for or a free one you found yourself. Contractors get this with their engagement and it forms part of it.
Approved tools today: [list them]. Work goes through the account the business provides, not a personal one, because that is the account we have checked and the one we can switch off. Paying for a plan is not the same as it being safe. Anything not on the list gets used for work only once [Name] has approved it.
Drafting, rewriting, summarising, brainstorming, formatting, formulas and code, and research you are going to verify. If it is not on the never list below, and you are on an approved tool and an approved account, go ahead. You do not need permission each time.
Passwords, API keys and access credentials. Bank, tax, Medicare, passport and licence numbers. Health information about an identifiable person. Anything a client contract or a confidentiality agreement forbids. Anything subject to legal professional privilege.
Client information goes only into approved tools, only for that client's own work. Where you can, describe the situation rather than pasting the document, and take the names out. Then check that what is left does not identify them anyway, because in a market this size the details usually do. If the question only works with the identifying detail in it, it is a question for a colleague.
Hiring, shortlisting, performance, pay and termination decisions are made by a person who can explain the reasons without reference to a tool. AI can help you prepare, sort or summarise. It does not choose, score or rank people.
Ask everyone in the meeting before you turn on a notetaker, every time, and stop if anyone says no. Recording a private conversation without the consent of everyone in it is an offence in New South Wales, and most AI notetakers join and start recording by default.
Nothing an AI tool produced goes to a client, a supplier, a regulator or the public until someone here has read it and is willing to put their own name to it. If AI answers customers directly, that is a decision [Name] makes, it says that it is AI, and someone reads a sample of what it said each week.
Numbers, dates, legal or regulatory statements, quotes, citations and anything about a named person get checked against the source. These tools produce wrong answers in the same confident tone as right ones, so confidence is not a signal you can use.
You do not need to announce that AI helped you draft an email. You do need to say so when a client has asked, when it would change how they read the work, or when it produced something they will rely on as your professional judgement. Where a court, a regulator or a client contract sets a disclosure rule, that rule wins.
If confidential information went somewhere it should not have, or AI output went out with a mistake in it, say so the same day. Do not delete the conversation: export or screenshot it first, because it is the only evidence of what went in. [Name] then decides within a day whether anyone outside the business needs to be told. Nobody is in trouble for reporting a mistake. Deliberately ignoring this policy is a different thing and is handled like any other breach of our workplace rules.
[Name] is responsible for this policy, keeps the approved tools list, and answers the questions it does not. Adopted [date]. Next review [date], and again whenever we add a tool.
A rule whose reason you cannot explain is one your team will route around the first time it is inconvenient.
The account is the thing you can check and switch off. A paid plan is not automatically a safe one, and framing the rule around cost sends someone with a personal paid subscription exactly the wrong way.
A policy that only lists what is forbidden gets read as a ban. People then either stop using the tools or guess, and guessing is what causes the problem the rest of the policy is trying to prevent.
"Do not share confidential information" is unarguable and useless, because everybody already believes they are following it. A list you can hold a document against is one people can apply at the moment they are deciding.
AI notetakers join and record by default, so nothing gets pasted anywhere and the rest of the policy never engages. For a firm that runs on client meetings this is the most likely breach and the least likely to be noticed.
Preserving the record comes before everything because the evidence disappears first. The amnesty is there because a team that hides a mistake for a fortnight turns a fixable problem into a serious one.
The work is in the four steps after you have written it.
Not a circulated document. Read clause 3 aloud and ask whether anything on that list has already gone into a chatbot. You will find out things no document would have told you.
The edges are where people get stuck, not the obvious cases. Take a real document from last week and decide as a group what could and could not be pasted.
Ask each person which account they use. This is the step that surfaces the personal logins, and it is better to find them now than in six months.
Alongside the other things they get on day one. A policy introduced later reads as a reaction to something, and people wonder what.
A policy that oversells itself is its own risk.
It is not legal advice. It is a sensible starting point written for ordinary Australian businesses. Two things catch professional services firms in particular: the small business exemption under the Privacy Act has been narrowed, and since 1 July 2026 accountants, lawyers, conveyancers and real estate businesses have been brought into the anti-money-laundering regime for the personal information they handle under those obligations. There is also a statutory tort for serious invasion of privacy, in force since June 2025, which does not care how small you are. If any of that is you, have your usual adviser read this before you rely on it. Whether a business your size needs formal governance at all is worked through in do Australian businesses still need AI governance.
It does not tell you how any particular tool handles your data. That changes by vendor, by plan and by month. Our platform pages go through what each assistant can reach and where the marketing overstates the difference.
It does not stop anything on its own. No policy does. What it does is make the expectation explicit, so a mistake becomes a conversation about a rule rather than an argument about what was reasonable.
If anyone on your team uses AI for work, you already have a policy. It is unwritten, different in each person's head, and invisible to you. Writing it down is how you find out that two people had opposite ideas about what was acceptable.
About a page. If yours runs much past that, it has usually started describing tools rather than behaviour, and the tools change faster than the document will. Length is the first reason people do not read them.
Yes, with two things to settle first: which tools you are approving, and whether anything in your client contracts, professional obligations or insurance already sets rules for this. It is a starting point rather than legal advice. If you handle health or credit information, or your practice falls under the anti-money-laundering obligations that were extended to accountants, lawyers, conveyancers and real estate businesses on 1 July 2026, have your usual adviser read it before you rely on it.
Everyone working in the account the business provides rather than a personal one, because that is the account you have checked and the one you can switch off. Paying for a plan is not the same thing as it being safe, and a tool nobody has checked is not approved just because someone put it on a card.
Do not delete anything yet. Export or screenshot the conversation first, because it is the only evidence of what actually went in and it is usually impossible to reconstruct afterwards. Then establish what went in, when, and on which account, and take advice on whether it is a notifiable breach. Then fix the cause rather than the person.
Every six months, and whenever you add a tool. Rules written for the tools you had last year quietly stop matching the ones you have now, and a policy naming a product you no longer use tells your team the whole document is out of date.
Book a discovery call and tell us how your team actually works. If a one-page policy is all you need, we will say so.
