Free AI policy template for Australian businesses (2026)
Free AI policy template for Australian businesses: a 12-section acceptable use policy, Copilot data rules, a tool register and where AI law stands in 2026.
An AI policy template gives an Australian business ready-made rules for how staff use ChatGPT, Microsoft 365 Copilot and other AI tools: which tools are approved, what information never goes into them, who checks the output and who signs off on anything new. Below is a free twelve-section template aligned to the National AI Centre's six practices, plus where Australian law stands in 2026.
Does your business need an AI policy?
If anyone on your team used an AI tool for work this month, yes. Without written rules, each person decides what is safe to paste into a chatbot. That is shadow AI, and it stays invisible until a client asks where their data went.
In its guidance on commercially available AI products, Australia's privacy regulator, the OAIC, says organisations should establish policies and procedures for using AI, train staff on its limits and, as best practice, keep personal information out of publicly available generative AI tools. A one-line ban rarely holds; a short, specific policy does.
What's law, what's guidance and what's coming?
Australia has no standalone AI law: existing laws apply, the national guidance is voluntary and privacy reforms are in draft. As at October 2026 (general information, not legal advice):
| Status | What it is | What it means for your policy |
|---|---|---|
| Law now | Privacy Act 1988 | Personal information typed into AI is covered, and the OAIC treats personal information that AI generates or infers as a collection. |
| Law now | Australian Consumer Law | A Treasury review found consumers get the same protections for AI products and services as for any others. |
| Law from 10 December 2026 | Automated decision-making transparency (APP 1.7 to 1.9) | Privacy policies must describe significant decisions about people that computer programs make or substantially and directly help make. |
| Voluntary guidance | Guidance for AI Adoption (National AI Centre, October 2025) | Six practices: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, maintain human control. |
| Policy direction | National AI Plan (2 December 2025) | Dropped the proposed mandatory guardrails for high-risk AI in favour of existing laws, more guidance and a new AI Safety Institute (Piper Alderman analysis). |
| Proposed, not law | Draft Privacy Amendment (Personal Data Protection) Bill 2026 | Would expressly treat AI-generated and derived data, including algorithmic inferences, as collection (Clayton Utz summary). |
The Guidance for AI Adoption is the first update of the 2024 Voluntary AI Safety Standard. Its Foundations version targets small and medium businesses and not-for-profits, and the guidance includes its own policy and register templates; ours is shorter and built around Microsoft 365. For the draft bill, read our 2026 Privacy Act reforms explainer.
What goes in an AI acceptable use policy?
Twelve sections cover what staff need and what the six practices expect. Use the model wording as your AI usage policy for staff; [square brackets] mark what to fill in.
Scope.
"This policy covers everyone working for [Business name], including contractors, using AI for work: chatbots, Copilot, AI features in other software, note-takers and browser agents."
Accountability.
"[Name, role] owns this policy and the AI tool register. Only they approve new tools or new uses." (Practice 1: decide who is accountable.)
Approved tools only.
"Beyond public information, use only tools on the register, signed in with your work account. Personal accounts, free versions and browser extensions are not approved for internal or client information."
Restricted information.
"Passwords, MFA codes and other secrets never go into any AI tool. Health information, tax file numbers, card or bank details and identity documents go only into tools the register approves for them."
Everyday uses.
"Approved AI may be used to draft, summarise, rewrite, brainstorm and analyse information you can already access."
Approval first.
"Get written approval before using AI to screen job applicants, score or rank customers, set individual prices, advise clients or act for you in email, files or websites." (Practice 2: understand impacts and plan accordingly.)
Human review.
"You own anything you send, file or decide, whoever drafted it. Check AI output for accuracy, bias and confidential detail before it reaches a client, patient or decision." (Practice 6: maintain human control.)
Openness.
"Customers are told when they are talking to an AI chatbot. Meeting attendees are told before an AI note-taker joins." (Practice 4: share essential information.)
Connections.
"Never connect AI apps or agents to company email, files or calendars yourself. [IT contact] checks vendor data-use and retention terms first."
Incidents.
"If you enter something you should not have, or AI output could harm someone, tell [contact] straight away. Reporting early matters more than blame." (Practice 3: measure and manage risks.)
Training.
"Everyone completes AI training at induction and whenever approved tools change."
Review.
"[Owner] reviews this policy and the register every [six] months, after any AI incident and when a tool changes how it handles data. Staff sign to confirm they have read it." (Practice 5: test and monitor.)
What can staff put into Copilot vs public AI tools?
Sort information into five classes and give each tool a yes or no. This table is the core of a Copilot usage policy; the last column assumes permissions are tidy.
| Information class | Examples | Free or personal AI tools | Copilot Chat (work account) | Microsoft 365 Copilot (licensed) |
|---|---|---|---|---|
| Public | Website copy, published prices, job ads | Yes | Yes | Yes |
| Internal | Procedures, meeting notes, draft proposals | No | Yes | Yes |
| Confidential | Client files, contracts, financials, staff records | No | Approved tasks only | Yes, within each person's access |
| Sensitive personal | Health information, tax file numbers, card or bank details, identity documents | Never | No | Only if the register approves it, labelled and DLP-covered |
| Secrets | Passwords, MFA codes, API keys | Never | Never | Never |
With a work account, Copilot Chat and Microsoft 365 Copilot (which Microsoft is renaming Microsoft Copilot, privacy terms unchanged) carry Microsoft's enterprise data protection: prompts and responses fall under the same Data Protection Addendum as your email and SharePoint files, and Microsoft says they are not used to train foundation models. Microsoft 365 Copilot only surfaces content each person can already view, so over-shared folders are the real exposure, and the SearchLeak Copilot flaw showed a hijacked Copilot inherits that reach. Add three settings to your appendix:
Web search.
Copilot sends Bing a few words from the prompt, stripped of user and tenant identifiers. Microsoft's web search documentation says the Data Protection Addendum does not cover those queries, though its Product Terms rule out advertising use and model training. Admins can turn web search off, or off in work mode only.
Labels and DLP.
Microsoft Purview DLP can stop Copilot using files and emails with chosen sensitivity labels and block web searches on prompts containing sensitive information, licences permitting. It cannot scan files uploaded into a prompt, so the "never" rows still rely on people.
Processing location.
Microsoft's Copilot privacy documentation says customers outside the EU may have queries processed in the US, the EU or other regions. Note it if a client contract requires Australian data residency.
What goes in an approved-tool register, and who signs off?
The National AI Centre recommends an AI register recording accountable people, purpose, capabilities and limitations, risks and their treatment, and review dates. For a small business, one row per tool is enough:
| Register field | Example entry |
|---|---|
| Tool and account type | Microsoft 365 Copilot, work accounts only |
| Business owner | Operations manager |
| Approved uses and information classes | Drafting and meeting summaries; public, internal and confidential |
| Settings and vendor terms checked | Web search off in work mode; DLP excludes the Highly Confidential label; no model training on prompts |
| Decisions about people? | No |
| Next review | [Date] |
Three roles sign off: the business owner (accountable), whoever handles privacy, and your IT provider (settings, vendor terms and app permissions). By default, Microsoft 365 users can let an app into their own mailbox without an administrator; restricting user consent and enabling admin consent requests sends new AI apps to the register first. Give AI browser agents their own row; they act inside a person's logged-in sessions.
What must you disclose about automated decisions from December 2026?
Starting 10 December 2026, any business bound by the Privacy Act has to say in its privacy policy when computer programs make, or substantially and directly help make, decisions that could significantly affect someone's rights or interests, and what kinds of personal information they use. The OAIC published a fact sheet and flowchart on 30 September 2026. Section 6 routes these uses through approval, and the register's "Decisions about people?" field gives your privacy policy its list. Our guide to automated decision disclosures covers the wording.
How do you put this template to work?
The twelve sections, information table and register above are the complete template: paste them onto your letterhead and fill the brackets. Prefer an editable Word version? Ask us for the Word copy. Then check your data with the free AI readiness assessment.
Need help rolling it out?
Through our AI governance work, Key IT's security-certified engineers tidy SharePoint and Teams permissions before Copilot goes live, set up labels and DLP where licences allow, restrict app consent and train staff on the yes-and-no table, across Melbourne and remotely Australia-wide. Call 1300 053 948.
Frequently asked questions
Is an AI policy a legal requirement in Australia?
No standalone AI law requires one, and the National AI Centre's guidance is voluntary. But the Privacy Act and consumer law already apply to staff use of AI, and the OAIC expects organisations to establish policies and procedures for AI systems. A signed policy is your evidence.
Can staff use ChatGPT at work?
Only through an account your business has approved and listed on the register. Before approving a chatbot, check the vendor's business terms: whether prompts train its models, how long they are kept and where they are processed. Personal and free accounts give you no say, so the template limits them to public information.
Is Microsoft 365 Copilot safe to use with client data?
It can be, once permissions are right. Microsoft says prompts, responses and data accessed through Microsoft Graph are not used to train foundation models, and Copilot only shows what each user can already open. If a client folder is open to everyone, Copilot will summarise it for anyone.
Does a small business need a generative AI policy?
Yes, arguably more than a large one, since small teams adopt tools fast. Many businesses under $3 million turnover sit outside the Privacy Act, but health service providers, businesses trading in personal information and Commonwealth contractors are covered at any size, and client contracts often carry their own confidentiality terms.
Does the draft Privacy Bill 2026 change our AI policy now?
Not yet. It is an exposure draft, and consultation closed on 18 September 2026. Its direction matters, though: AI-generated and derived data, including inferences about people, would expressly count as collection. The OAIC already treats AI-generated or inferred personal information that way, so a policy written to that standard now should hold up.
What should staff do if they paste something sensitive into an AI tool?
Tell the policy owner straight away, noting what was entered, into which tool and from which account. The owner can delete the chat where the tool allows, check the vendor's retention terms and decide whether to assess it as a possible eligible data breach. A no-blame rule gets you told early.
Book your free IT & Cyber Security Review
See exactly where your IT and security stand, and what to fix first. No jargon, no obligation.