Privacy and data
How to write a workplace AI use policy
A workplace AI use policy tells staff which AI tools they may use, what they must never enter, who checks the output and what happens if the rules are broken. The OAIC says organisations should establish policies and procedures for AI use, and backing them with training is what makes them work.
Updated . General information, not legal advice.
Do I need a workplace AI use policy?
If staff use AI tools at work, yes. The OAIC says organisations should establish policies and procedures for the use of AI systems, and APP 11 requires reasonable steps, including organisational measures, to protect personal information.
Most staff already have access to AI features in email, document and meeting software, whether or not you have approved them. Without a policy, nobody knows whether pasting a customer complaint or a medical certificate into a free chat tool is allowed. A short written policy sets the line, gives managers something to train against and gives you a fair basis for dealing with misuse.
Even if your business is under the Privacy Act’s $3 million turnover threshold, the tax file number rules, confidentiality obligations to customers and discrimination law still apply. For the legal background, see AI for HR in Australian small businesses.
What should an AI use policy cover?
Ten topics: purpose and scope, approved tools, banned inputs, human review, decisions about people, confidentiality and IP, telling customers, training, breaches and consequences, and review.
Keep the policy short enough that people read it, ideally two pages. Put the detailed tool list in an appendix you can update without reissuing the policy. The outline at the end of this guide gives wording topics for each section.
How should the policy deal with approved tools?
Name the tools staff may use, the accounts they must use them through, and the kinds of information each tool is approved for. Anything not on the list is not approved for work.
- Approve business accounts the organisation controls, not personal accounts, so you can manage access and remove it when someone leaves.
- Record each tool’s settings: whether inputs are used for training (the OAIC flags this as a disclosure risk), where data is stored, and how long it is kept.
- Give each tool an approval level, such as ‘no personal information’, ‘staff information without sensitive details’ or ‘approved for HR records’.
- Name an owner who approves new tools and reviews existing ones. The OAIC says AI tools should be reviewed regularly, not set and forgotten.
What information should staff never enter into an AI tool?
Tax file numbers, bank and card details, health and other sensitive information, passwords and confidential customer information should be banned from any tool not specifically approved for them.
| Never enter | Why |
|---|---|
| Tax file numbers | Anyone holding TFN information is a file number recipient and must follow the TFN rules (Privacy Act s 11 and s 18). |
| Health information, including medical certificates and injury details | Sensitive information under s 6(1) of the Privacy Act. |
| Union membership, racial or ethnic origin, religion, sexual orientation, criminal record, biometric data | Also sensitive information under s 6(1). |
| Bank account and card details | Exposure can enable payment fraud and may be a notifiable data breach. See Essential Eight for HR data. |
| Passwords, access codes and system keys | Gives anyone who sees the prompt access to your systems. |
| Confidential customer, supplier or commercial information | May breach contracts or confidentiality duties. |
As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools. Tell staff to remove names and identifiers when the task doesn’t need them.
What should the policy say about checking output and decisions about people?
Require a person to check facts, figures and tone before any output is used, and ban AI tools from making the final decision on hiring, discipline, pay or dismissal.
AI tools produce confident errors. Under APP 10, organisations covered by the Privacy Act must take reasonable steps to make sure personal information they use and disclose is accurate, and the OAIC says information an AI tool produces about a person is personal information even when it is wrong. The person who sends or relies on the work stays responsible for it.
Decisions about people carry the most risk. The Fair Work Act bans adverse action against employees and prospective employees because of attributes such as age, sex, race or disability (s 351). See using AI in recruitment for screening tools.
How should the policy handle confidentiality and intellectual property?
Under section 35(6) of the Copyright Act 1968, the employer owns any copyright in literary, dramatic or artistic work an employee makes under their employment. The policy should keep confidential material out of unapproved tools and require permission before entering others’ material.
- Work made using an approved tool is still work for the business: say it belongs to the business and is handled like any other work document.
- Don’t enter material owned by a customer, supplier or other third party unless the contract or licence allows it.
- Don’t present AI-assisted work as independently checked or as someone’s own expert opinion when it isn’t.
- Keep AI-assisted drafts of contracts, advice or published content under the same review as anything else you publish.
Do we need to tell customers when we use AI?
The OAIC says public-facing AI tools such as chatbots should be clearly identified as AI, and privacy policies should explain AI use. From 10 December 2026, APP 1.7 requires privacy policies to cover programs that make or substantially help make significant decisions.
Put the customer rule in the policy: who may switch on a customer-facing AI feature, how it is labelled, and who updates the privacy policy and collection notices. If a program shapes decisions that significantly affect customers or applicants, see automated decisions and privacy disclosure.
What about training, breaches, consequences and consultation?
Train staff before they get access, make reporting a mistake easy, deal with breaches through a fair process, and consult employees where an award or agreement consultation term applies to the change.
- Training: the OAIC says any ban on entering personal information needs to be backed by training and auditing. Cover the banned inputs and the approved tools list at induction and at each policy change.
- Reporting: tell staff to report straight away if they enter something they shouldn’t have. A covered business must take reasonable steps to assess a suspected eligible data breach within 30 days (Privacy Act s 26WH).
- Consequences: say that breaches will be dealt with through a fair process and can lead to disciplinary action. Distinguish honest mistakes reported promptly from deliberate misuse.
- Consultation: award consultation clauses generally apply when an employer has decided to introduce major changes in technology likely to significantly affect employees. The Fair Work Ombudsman says you must notify and discuss with affected staff, give written information and consider their views, but consent isn’t required.
If an AI feature monitors or records staff emails, chats or activity, state surveillance laws may require notice first. See using AI assistants with HR data.
What does a section-by-section outline look like?
Use ten short sections in this order, and put the approved tools list in an appendix you can update without reissuing the policy.
| Section | What to write |
|---|---|
| 1. Purpose and scope | Who the policy covers (employees, contractors, volunteers) and which tools: chat tools, AI features in other software, browser extensions. |
| 2. Approved tools | Point to the appendix. Only approved tools, through business accounts, for work. |
| 3. Banned inputs | The table above, in plain words. Remove identifiers when not needed. |
| 4. Checking output | A person checks facts, figures, tone and sources before use. You stay responsible for your work. |
| 5. Decisions about people | No AI tool makes the final decision on hiring, discipline, pay or dismissal. Record when a tool was used. |
| 6. Confidentiality and IP | Work belongs to the business. No third-party material without permission. |
| 7. Customers | Who approves customer-facing AI, how it is labelled, privacy policy updates. |
| 8. Training and support | Training before access. Who to ask. |
| 9. Breaches and consequences | Report mistakes immediately. Fair process for misuse. |
| 10. Review | Owner and review date, such as every 6 or 12 months, and after any incident. |
The HR policy builder includes a technology and AI tools policy you can use as the core of sections 2 to 4 and adapt with the rest of this outline.
Questions
Is a workplace AI use policy required by law?
No law we checked requires a stand-alone AI policy by name. The OAIC says organisations should establish policies and procedures for AI use, and a written policy is one of the organisational measures APP 11 contemplates for covered businesses.
Should the policy cover AI features built into software we already use?
Yes. Summaries in email, meeting transcription and drafting features in document software handle the same information as a stand-alone chat tool, so list them in the approved tools appendix with their approval level.
Do we need to consult staff before introducing the policy?
If the AI rollout is a major change likely to significantly affect employees, award consultation clauses and the consultation term in an enterprise agreement generally apply. Check the clause that covers your staff.
Sources
- OAIC: Guidance on privacy and the use of commercially available AI products
- Privacy Act 1988 (Cth), current compilation (ss 6(1), 11, 18, 26WH; APPs 10, 11)
- Privacy and Other Legislation Amendment Act 2024, s 2 and Schedule 1 Part 15
- Copyright Act 1968 (Cth), current compilation (s 35(6))
- Fair Work Act 2009 (Cth), current compilation (s 351)
- Fair Work Ombudsman: Consultation and cooperation in the workplace
Checked against these sources on 12 November 2026. Spotted something out of date? Tell us.