Free security resource · No email required · Plain English

Stop client data leaking into consumer AI tools.
A usage policy your staff will actually read. Copy it, adapt it, adopt it.

Somebody in your practice has already pasted a client email, a tax return or an ID document into a free chatbot to save ten minutes. This policy is a page and a half that tells honest people where the line is. Written for Australian practices, free, no email gate. The highlighted fields are yours to fill in.

Or start with the 10-question self-check: is your team using AI safely? ↓

Before you adopt it

We treat staff use of consumer AI as a data-leak problem, not an innovation project. A policy on its own does not stop data leaving: it tells honest people what right looks like. Pair it with the technical controls in our managed cyber security baseline (Purview data loss prevention, Defender for Cloud Apps to spot unapproved tools, Conditional Access) and it becomes real. This template is general information, not legal advice; if you operate under specific regulatory obligations, have your adviser review the adapted version.

Client data stays put

TFNs, ID, bank details and advice records never go into a tool you have not approved.

Australian-hosted first

If a tool is approved for client information, prefer one that keeps that data in Australia.

Evidence on request

A dated, owned policy is one of the first things a licensee, insurer or auditor asks to see. AFS licensees: see what ASIC found in REP 798.

The template

AI usage policy

[BUSINESS NAME] · AI USAGE POLICY · VERSION 1.0 · ADOPTED [DATE] · OWNER [NAME / ROLE] · REVIEW EVERY 6 MONTHS

1. Why this policy exists

AI tools make it easy to send business and client information to systems we don't control, often overseas. Our clients trust us with that information. This policy exists so everyone knows which tools are approved, what can and can't go into them, and what to do when something goes wrong. It applies to everyone who works for [BUSINESS NAME], including contractors, on any device used for work.

2. Approved tools

The following AI tools are approved for work use, on business accounts only:

  • [TOOL NAME, PLAN, and whether client information is permitted in it]
  • [Add or remove tools. "None" is a valid answer.]

Where a tool is approved for client information, it must keep that information in Australia and must not use it to train its models. [OWNER] confirms both before a tool is added.

Anything not on this list is not approved for work information. If a tool would help you, ask [OWNER]. Asking takes a day, not a committee.

3. Accounts

  • Work use of AI happens on work accounts. Never use a personal AI account for work information, even once, even if it's faster.
  • Sign in with your work identity (single sign-on) where it's offered.
  • Don't create new AI accounts with your work email without approval. Approved tools are set up for you.

4. What must never go into any AI tool

  • Passwords, access keys or anything used to log in to anything.
  • Payment card numbers, bank account details, or government identifiers (TFNs, Medicare numbers, licence numbers).
  • Health information about any identifiable person.
  • Anything a client has told us in confidence that identifies them, unless the tool is on the approved list and [OWNER] has confirmed client data is permitted in it.
  • Anything covered by a specific confidentiality obligation (contracts, disputes, sensitive HR matters) without approval.
Rule of thumb: if you'd hesitate to email it to an outside supplier, don't paste it into an AI tool without checking.

5. Using AI output

  • You are responsible for what you send, publish or act on. "The AI wrote it" is not a review process.
  • Check facts, figures, names and anything with legal, financial or safety consequences before it leaves the business.
  • Don't present AI-generated work to a client as something it isn't. Where disclosure matters in our industry, follow [INDUSTRY-SPECIFIC RULE, IF ANY].

6. When something goes wrong

If work or client information ends up in an unapproved tool or a personal account, yours or anyone's, tell [OWNER] the same day. The goal is fixing it fast, not blame; the only career-limiting move is hiding it.

7. Keeping this policy honest

  • [OWNER] reviews this policy and the approved-tools list every 6 months, or when a new tool is adopted.
  • New starters get this policy in week one, with a 15-minute walkthrough of the approved tools.
  • Questions and "can I use X?" requests go to [OWNER / CHANNEL].

LICENCE: COPY AND ADAPT THIS FREELY FOR YOUR OWN BUSINESS. NO ATTRIBUTION NEEDED. GENERAL INFORMATION, NOT LEGAL ADVICE.

Self-check · 10 questions · No email required

Is your team using AI safely?

Tick each statement that is true today, then count the ticks. Nothing is sent anywhere: this is a list on a page, and it works on paper too.

8 to 10

You are ahead of most. Review it every six months, and when a new tool arrives.

5 to 7

The policy above closes most of the gaps. Adopt it, name an owner and list the tools.

4 or fewer

Staff are deciding for themselves. Start with the policy, then find out what is really in use.

The third statement is the hard one, because people do not volunteer it. In an Automation & AI Discovery we find out through confidential 1-on-1s. Want the pre-work questionnaire we send to staff first? Ask for it and an engineer will email it to you.

↳ The policy is the easy half

Want it enforced by your tenant, not just adopted by your staff?

Our managed cyber security baseline sets the controls this policy assumes: data loss prevention that notices TFNs and ID documents, detection of unapproved AI tools, and sign-in rules. The Live Audit shows you where your tenant stands today.