AI and the Privacy Act: what must an Australian business check?
Seven checks before you paste.

The quickest way to use an AI tool is to paste in a customer email and ask for a reply. That puts personal information into someone else’s software. Here is who the Act covers and what the Office of the Australian Information Commissioner (OAIC) expects.

We sell AI automation, so read this as general information, not legal advice. Every legal point names its source at the foot of the page.

The short answer.

First check whether the Privacy Act 1988 covers you. A business with annual turnover of $3 million or less is generally exempt, but health services, businesses that trade in personal information and some others are covered whatever their size. If it does, do due diligence on the product. Update your privacy policy and notices to explain your AI use. As best practice, keep personal information, particularly sensitive information, out of public AI tools. Treat what the AI produces about a person as personal information, with a person checking it.

Does the Privacy Act apply to my business?

This guide covers the federal Privacy Act 1988 only. State and territory laws can add their own rules.

The Act generally covers businesses with annual turnover above $3 million. A business at or under that figure is usually exempt as a small business operator, but one that went over $3 million in an earlier financial year while it was running the business generally stays covered even if its turnover has since dropped. A business not carried on at all last financial year is small only if its current-year turnover is $3 million or less (section 6D(2)). A new owner of an existing business is judged on the years they have run it.

Under section 6D, a small business is still covered if it provides a health service and holds health information other than in an employee record. The same goes for one that discloses personal information for a benefit, service or advantage, or provides a benefit, service or advantage to collect it, unless every person concerned has consented or legislation requires or authorises it. Commonwealth contracted service providers and credit reporting bodies are covered too.

What counts as personal information in AI?

The Act defines personal information as information or an opinion about an identified individual, or one who is reasonably identifiable, whether it is true or not. In an AI tool, that means inputs and outputs.

What you put in

Pasting an enquiry or a call transcript into an AI tool puts personal information into it if the text contains information about an identified or reasonably identifiable person.

The OAIC urges extra caution with sensitive information, which includes health information, racial or ethnic origin, religious beliefs, sexual orientation and criminal record.

What comes back

The OAIC says information an AI product infers or generates about an identifiable person is personal information, even when it is wrong, including hallucinations and deepfakes. So an invented detail in an AI-written customer summary is still personal information and, if the Act covers you, falls under the Australian Privacy Principles (APPs).

What should I check before using an AI tool?

The OAIC’s guidance on commercially available AI products, published 21 October 2024, covers products a business buys and free public tools such as chatbots.

Privacy checks before using an AI product, with what the OAIC guidance says and what to do
CheckWhat the OAIC guidance says What to do
Is the product fit for this job?Do due diligence: testing for your intended use, human oversight, privacy and security risks, and who can access the informationName the one job it will do and ask the vendor for evidence it suits that job
Can the vendor use your data?Check whether the terms give the developer access to your inputs and outputs, including for further trainingRead the terms and switch off training or sharing features where you can
Does your privacy policy mention AI?Privacy policies and notices should give clear, transparent information about your use of AIAdd a plain paragraph on what the AI does with customer details and who provides it
Will staff paste personal details into a public tool?As best practice, do not enter personal information, “particularly sensitive information”, into publicly available generative AI toolsWrite a staff rule on what must never go into a public tool, and train staff on it
Is the output about a person?Generated or inferred information about an identifiable person, even if incorrect, is personal informationCheck that you may collect it (APP 3). If not, destroy or de-identify it as soon as practicable, where that is lawful and reasonable and it is not in a Commonwealth record (APP 4.3). If you may keep it, treat it like any customer record and correct what is wrong
Who checks the output?Build appropriate human oversight of AI outputs into your processesName the person who approves anything before it is sent or saved
Is anyone reviewing it later?Due diligence should not be “set and forget”; review and monitor across the product’s lifeDiary a review and repeat these checks when the terms change

From 10 December 2026, if the Act covers you and a computer program uses personal information to make a decision, or do something substantially and directly related to making one, that could reasonably be expected to significantly affect a person’s rights or interests, your privacy policy must describe the kinds of personal information used and, separately, the kinds of decisions the program makes on its own and the kinds where it does something substantially and directly related to making the decision (APP 1.7 to 1.9).

To pick the first job for AI, use the AI automation playbook.

Where does the AI provider keep the data?

Ask where your data is processed and stored, and who can see it, before anything personal goes in.

Overseas disclosure (APP 8)

APP 8 sets rules for disclosing personal information to a recipient outside Australia. If a provider handles your customers’ details offshore, ask it to name the countries in writing.

Security (APP 11)

APP 11 means taking reasonable steps to protect personal information from misuse, interference and loss, and from unauthorised access, modification or disclosure. For an AI tool, the OAIC suggests searching for past security incidents and checking the developer’s security measures.

Notifiable data breaches

An eligible data breach is unauthorised access or disclosure that a reasonable person would conclude is likely to result in serious harm. For lost information, the test is whether unauthorised access or disclosure is likely and, if it happened, whether serious harm would be likely. If you have reasonable grounds to suspect one, you must assess it quickly and take all reasonable steps to finish the assessment within 30 days. If you have reasonable grounds to believe one has happened, you must prepare a statement (your identity and contact details, what happened, the kinds of information involved and the steps you recommend) and give a copy to the Information Commissioner as soon as practicable (section 26WK). Then, as soon as practicable, tell the people affected what it says, either everyone involved or those at risk (section 26WL). If neither is practicable, publish the statement on your website, if you have one, and take reasonable steps to publicise it. Remedial action can change this (section 26WF), but only if you act before serious harm results and a reasonable person would then conclude it is not likely. Then there is no eligible data breach, as with a sensitive file sent to the wrong person and confirmed deleted unread. If the test is met for only some people, they need no notice, but the Commissioner and anyone still at risk must still be told.

Voice tools raise further questions, covered in our guide to whether an AI receptionist is legal in Australia.

Where Pixelate fits.

BISMO, our lead AI agent, prepares replies, bookings, invoices and payment chasers inside your own CRM, and by default nothing goes to a customer until you approve it. See our AI automation page. Put the checks above to us in writing, and compare our answers with other providers’ using our guide to choosing an AI automation provider.

Questions owners ask about AI and privacy.

  • Is a free chatbot account safe for customer details?

    The OAIC recommends, as best practice, that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools, because of the significant and complex privacy risks. A paid business version may let you limit what the vendor can access or train on, but the OAIC’s due diligence checks still apply before any personal information goes in.

  • Do I need to tell customers I use AI?

    If the Privacy Act covers you, the OAIC says privacy policies and notices should give clear, transparent information about your use of AI, and public-facing AI tools such as chatbots should be identified as AI to the people using them.

  • Can AI make decisions about customers on its own?

    The OAIC guidance expects appropriate human oversight of AI outputs, with a person responsible for checking the accuracy of personal information the AI produces and able to overturn its decisions. In practice, name who approves anything the AI sends, saves or decides before it takes effect.

  • We turn over less than $3 million. Can we ignore this?

    Not always. A carve-out can remove the exemption, such as providing a health service while holding health information, or trading in personal information outside the Act’s exceptions. A business that went over $3 million in an earlier financial year while it ran the business can also stay covered. Even if you are exempt, the OAIC checks are a sound standard for customer details in AI.

Sources.

Read and checked on 12 November 2026. How we research and check guides is set out in how we write our guides.

Ask us the seven checks.

Tell us the job. We will answer in writing: what customer information it touches, who approves what it sends, and what it costs.