Privacy Act AI tools Australia: does it apply to your business?
The Privacy Act 1988 and the Australian Privacy Principles do not apply to every Australian business. Many small businesses below an annual turnover threshold are generally outside the Act, but there are significant exceptions, including some health service providers and businesses that trade in personal information. The rules have also been under review, so the detail can change. Check the current position for your business rather than relying on a general article, including this one.
Even if the Act does not apply to you, other things may. Client contracts often include confidentiality and data handling terms. Professional obligations may add their own. Your customers will usually expect care regardless of whether you are technically covered. And treating personal information carefully is simply sound practice.
This guide takes the practical view: assume you want to handle personal information well, then check which formal obligations sit on top.
One more reason to check: if you supply a larger organisation or government, you may be bound by privacy terms in your contract even where the Act does not apply to you directly.
What counts as personal information in a prompt
Personal information is, in broad terms, information about an identified or reasonably identifiable person. Names and email addresses are obvious. Less obvious are a file note describing a client's situation with no name but enough detail to identify them, a photo, a call recording, a transcript, a CV, a rental application, a clinical note or a complaint.
Some information is classed as sensitive and generally attracts stricter treatment: health information is the one many businesses meet, along with things like racial or ethnic origin, criminal records and union membership. If your staff handle any of this, the bar for putting it in an AI tool is much higher.
A useful test for staff: if this text turned up on a screen in front of the person it describes, would they be surprised that it had been given to this tool?
Remember that de-identifying is harder than deleting a name. A paragraph that mentions a rare job title, a small town and a specific incident may still point to one person. Ask whether someone who knows the situation could tell who it is about.
Using an AI tool is often a disclosure to someone else
When a staff member enters personal information into an AI service, the information usually travels to the vendor's systems. In privacy terms that can be a use or disclosure of the information, and the Australian Privacy Principles set expectations about using information for the purpose it was collected, or a related purpose people would reasonably expect.
Three other themes are worth knowing by name. Overseas disclosure: many AI services process data outside Australia, and the principles address sending personal information overseas. Security: businesses are expected to take reasonable steps to protect personal information, and to destroy or de-identify it when it is no longer needed. Collection: some AI features create new information about people, such as transcripts or summaries, which is itself a form of collection.
Whether a specific use is allowed turns on the facts: what the information is, what people were told, what the tool does with it, and where. That is the point at which to get advice rather than guess.
When a notetaker produces a transcript, or a tool scores a CV, the business now holds new personal information it did not have before. That information needs the same care as anything else: who can see it, how long it is kept, and whether the person would expect it to exist.
Questions to ask any AI vendor before personal information goes in
A free or consumer account often carries weaker commitments than a business one. That is a common reason to insist on business accounts for anything involving other people.
Put these to the vendor or read the answers in their terms and trust documentation. If you cannot find an answer, treat that as an answer.
- Where is our data stored and processed, and by whom?
- Is what we enter used to train or improve models, and can that be switched off at the account level?
- How long is it kept, and can we delete it?
- Who at the vendor can see it, and under what circumstances?
- Does the business plan offer admin controls, access logs and single sign-on that a personal plan does not?
- What do the terms say about notifying us if there is a breach?
Practical rules for staff
Examples help more than rules. Show staff a before and after: a draft email to a tenant about arrears with the name, address and amount removed, and then the same email with the details put back in by the sender. The tool sees the first version; the human adds the second.
Staff do not need a lecture on the Australian Privacy Principles. They need rules short enough to remember in the middle of a busy afternoon.
- Use only approved business accounts for anything involving other people's information.
- Remove names and identifiers where you can, and ask whether the task works without them.
- Never enter sensitive information, such as health details, unless that use has been specifically approved.
- Check any output about a real person before it goes anywhere. AI tools state invented things confidently.
- Do not rely on a tool alone for decisions that affect an individual, such as hiring, credit, tenancy or care. A person decides.
- If you think you have gone too far, say so immediately. Quick reporting is welcome.
This guide describes privacy obligations in general terms. It is not legal advice, and the application of privacy law depends on your business, the information involved and current legislation. Take advice from a privacy lawyer or your industry body for your own situation.
When a mistake might be a data breach
Under the Notifiable Data Breaches scheme, organisations covered by the Act must notify affected individuals and the regulator when there is an eligible data breach, meaning unauthorised access or disclosure likely to result in serious harm. A staff member putting information into an unapproved AI tool is not automatically an eligible data breach. It may be, depending on what was entered, who could access it and what the harm could be.
The practical steps are the same either way. Contain it: stop further entries and, where possible, ask the vendor to delete. Assess it promptly: what, whose, how sensitive, who could see it. Record what you found and decided. Take advice early if the information was sensitive or the numbers are large.
Calibre Forum's survey can show you which tools people say they use, so you know where to look first. It does not monitor anything or advise on breach response.
Do not forget the contractual side. Some client agreements require you to tell the client about a data incident within a set time regardless of what the Act requires. Know what yours say.
A note on New Zealand
If you operate in New Zealand, the Privacy Act 2020 is the starting point. It works through information privacy principles, applies broadly to organisations that hold personal information, and is overseen by the Office of the Privacy Commissioner. It includes a requirement to notify privacy breaches that cause, or are likely to cause, serious harm, and it has rules about disclosing personal information to overseas parties.
The differences from Australia matter in detail, especially around coverage of small businesses, breach notification and overseas disclosure. For a business working on both sides of the Tasman, a sensible default is to apply the stricter expectation of the two to your staff rules, then get local advice on the formal position in each country.
If you serve New Zealand customers from Australia, or the reverse, check whether the other country's law applies to that activity as well.
Common questions
Does the Privacy Act apply to small businesses?
Often not, but sometimes yes. Many small businesses below an annual turnover threshold are generally exempt, though there are exceptions, such as some health service providers and businesses that trade in personal information. Because the law has been under review, check the current position for your business with a privacy lawyer or the Office of the Australian Information Commissioner.
Is it a breach to put client names into ChatGPT?
Not automatically, but it may be a use or disclosure that needs a proper basis. It depends on what the information is, what the client expects, the tool's terms and where data is processed. Entering sensitive information into a personal account is a far greater concern than a client's name in a business account with training switched off.
Do I need to tell customers that I use AI tools?
There is no single answer. If privacy law applies to you, your privacy policy and collection notices should describe how you handle personal information, which may include AI tools and overseas processing. Your contracts or professional rules may also require disclosure. Check your obligations, and be honest if a customer asks.
How is New Zealand different for AI and privacy?
New Zealand's Privacy Act 2020 uses information privacy principles, covers organisations broadly, requires notification of privacy breaches that cause or are likely to cause serious harm, and has rules on sending personal information overseas. The details differ from Australia's, so businesses on both sides of the Tasman should check each country separately.
This guide is general information, not legal, insurance or financial advice. Check your own obligations with a qualified adviser.