AI privacy for small business in Australia: a plain guide

What should you keep out of public AI tools?

Start here, whatever the law says about your business. The privacy regulator, the Office of the Australian Information Commissioner (OAIC), recommends as best practice that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools.

In practice, do not paste in:

• Customer names, phone numbers, emails or addresses
• Health details, or notes about a person's circumstances
• Staff records, pay details or tax file numbers
• Bank or card details, passwords or login codes
• Documents you have promised to keep confidential

You can still get help with the task. Describe the situation without the person. Write 'a customer' instead of the name.

What comes out counts too. The OAIC says AI output about an identifiable person is personal information, even when it is wrong.

Does the Privacy Act apply to your business?

The OAIC's small business page says most small businesses are not covered by the Privacy Act, but some are. It defines a small business as one with annual turnover of $3 million or less.

Regardless of turnover, the OAIC says the Act covers a business that is, among other things:

• A health service provider. The OAIC's checklist includes allied health professionals, complementary therapists and child care centres
• Trading in personal information, such as selling or swapping a customer list
• A contractor providing services under a Commonwealth contract
• Related to a business that the Act covers
• A business that has opted in to the Act

There are more categories on the OAIC page, and its checklist also asks whether turnover has exceeded $3 million in any financial year since 2002. Work through the checklist yourself. It is not a substitute for legal advice, and neither is this article.

What do the Australian Privacy Principles ask for, in plain words?

If the Act covers you, you must follow the Australian Privacy Principles, known as the APPs. The OAIC's AI guidance points to a few that matter most when you use AI.

• Be open. Have a clear, current privacy policy, and include your use of AI. This is APP 1
• Collect only what you need, by fair means. Using AI to generate or infer personal information counts as collecting it. This is APP 3
• Use information for the reason you collected it. Putting a customer's details into an AI tool needs to match that purpose, be something the person would reasonably expect, or have their consent. This is APP 6
• Keep it secure from misuse, loss and unauthorised access. This is APP 11

If the Act does not cover you, the OAIC still recommends protecting any personal information you hold as best practice.

What changes on 10 December 2026?

A new transparency rule starts on 10 December 2026. It sits in APP 1.7 to 1.9 and applies to APP entities, meaning organisations the Privacy Act covers.

The OAIC explains that a covered business must add information to its privacy policy when all three of these are true:

• It has arranged for a computer program to make a decision, or to do something substantially and directly related to making a decision
• The decision could reasonably be expected to significantly affect a person's rights or interests
• Personal information about the person is used by the program

The policy must then describe the kinds of personal information used, the kinds of decisions made solely by the program, and the kinds of decisions the program substantially helps with.

This is a rule about telling people. It does not ban automated decisions. If the Privacy Act does not cover your business, this rule does not apply to you directly.

Does a chatbot or AI tool trigger the new rule?

It depends on what the tool decides, not on whether it is called AI. The OAIC's updated APP 1 guidelines say 'computer program' is meant broadly. It includes simple rule-based software as well as AI, and chatbots are named.

The test is the decision. The guidelines list rights and interests that would generally be in scope, including access to healthcare, housing, employment opportunities, education, financial assistance, and essential banking and credit. A program can be caught even when it only recommends.

So, for example:

• A website assistant that answers questions about opening hours is not making a decision of that kind
• Software that scores rental applicants, screens job applications or approves credit, using personal information, may well be

Those two examples are our reading, not a ruling. If a covered business uses software to help decide who gets a job, a tenancy or credit, get advice before 10 December 2026.

What does the ACSC tell small businesses to do?

The Australian Cyber Security Centre (ACSC) guide, Artificial intelligence for small business, is short and practical. Its key actions are:

• Set rules for how AI tools can be used and what information can be shared with them
• Review an AI vendor's privacy, security and data handling before you sign up
• Check AI outputs, and keep a person involved in important decisions
• Train staff on safe use, including protecting sensitive information

Its checklist adds questions for any vendor. What data does the tool collect, and where is it stored? Who owns it? Will it be used to train AI models?

The guide notes some providers may use submitted data for training, depending on settings or subscription. Our ChatGPT guide shows where to look.

A worked example: a physiotherapy clinic

For example, take a small physiotherapy clinic with turnover well under $3 million. This is an illustration, not legal advice for any clinic.

Is it covered? The OAIC lists health service providers as covered regardless of turnover, and names allied health professionals. So the clinic should assume the Privacy Act applies and confirm with its adviser.

The receptionist wants to use a public AI assistant to tidy up a letter to a patient about a treatment plan. That would put sensitive health information into a public tool, against the OAIC's best practice advice. A safer path is to ask for a general template with no patient details, then fill it in inside the clinic's own practice software.

The clinic also wants a website chat assistant. It limits the assistant to hours, services and booking steps. A notice says it is an AI and asks people not to share health details. The clinic adds the tool to its privacy policy.

When should you not use AI at all?

Sometimes the safe answer is to leave AI out.

• When the task cannot be done without sensitive information, and the tool has not been approved for it
• When you cannot find out where the data goes or whether it trains a model
• When the output would decide something important about a person and nobody will review it
• When you have promised a client confidentiality and have not checked that the tool fits that promise

For everything else, one page of rules is enough for most small teams. List the tools staff may use and what never goes in. Name who checks output and who to tell if something goes wrong.

For help with the practical setup, see our AI automation service or contact us. For legal questions about your obligations, speak to a lawyer.

Common questions