Australian Privacy Act and AI: Does a Small Business Need to Comply?

Often no, but not automatically. The Australian Privacy Act generally does not cover a business with annual turnover of $3 million or less, yet health service providers, businesses that trade in personal information, government contractors and several other types are covered whatever their size. Check those exceptions first, then check the AI tools.

The exemption is a statement about your business, not about the tool. A bakery with $900,000 turnover that pastes a customer list into a chatbot is probably outside the Act. A $400,000 childcare centre doing the same thing is probably inside it, because the OAIC lists child care centres among health service providers. What follows is general information, not legal advice: confirm your position on the OAIC site or with a qualified adviser.

Follow me on Instagram@sagnikteaches

The turnover test, and what counts as turnover

The OAIC defines a small business as one with annual turnover of $3 million or less. Turnover means all income from all sources in a financial year. It excludes assets you hold, capital gains and the proceeds of selling capital items. Costs are not subtracted, which surprises owners who think in profit terms: a catering firm with $3.4 million of sales and a 2% margin is over the line.

Connect on LinkedInSagnik Bhattacharya

The OAIC's own checklist phrases the question as whether turnover exceeded $3,000,000 in any financial year since 2002. That is the part owners miss. A business that grew past the threshold once, then shrank, is not treated as small. A business under a full year old projects a full year of income from the months it has traded.

Subscribe on YouTube@sagnikteaches

Write the answer down in one line, with the year and the figure you used. If your turnover sits within 15% of the threshold, treat yourself as covered for planning purposes, because next year's accounts will decide it and rebuilding habits afterwards is harder than setting them now.

Exceptions that catch small firms anyway

The OAIC small business guidance lists the situations where a business under $3 million is still covered. Read the list slowly, because several of them describe ordinary trades.

One more point from the OAIC: it names the Tax File Number Rule among the specific obligations some small businesses must meet, so a business that collects tax file numbers for payroll should read that guidance and keep those numbers out of AI tools altogether.

Four short examples of the exceptions in use

A care agency with 12 staff and $1.8 million turnover provides personal care to clients at home. Care that maintains a client's physical health, recorded in care notes, fits the OAIC's description of a health service, and the OAIC lists aged care and disability services that handle health information among its examples. Treat the agency as covered regardless of turnover. Its rosters, care notes and incident reports are personal information, and many of them are sensitive information (health).

A music school with $650,000 turnover teaches children piano and guitar. Teaching is not a health service, so the school looks exempt. If it also passes its parent email list to an instrument retailer in return for a payment or a discount, without every parent's consent, it could be trading in personal information and so covered.

A veterinary practice treats animals. The OAIC's health service examples concern services to individuals, so a vet's coverage is worth checking with an adviser rather than assuming either way. The owners' names, addresses and payment details are still personal information, and many vets also belong to larger groups, which brings in the related-company test.

A two-person bookkeeping firm with $380,000 turnover, whose clients are small companies, looks exempt on turnover. If it is a reporting entity under anti-money-laundering law or related to a bigger accounting group, the answer may change, and any clients' tax file numbers it holds bring in the Tax File Number Rule guidance. This is exactly where the OAIC's own checklist earns its ten minutes.

Why AI use changes the question, even for exempt businesses

Exempt does not mean free of risk. Three things sit outside the Act's coverage test.

• Contracts with customers and suppliers. A larger customer may require you to follow the privacy principles in a services agreement. If you are a subcontractor in a government-funded programme, the contract carries the obligation.
• The vendor's terms. Free and consumer AI plans can use your chats to improve models unless you switch that off. Business plans such as ChatGPT Business and Claude Team do not train on business content by default, according to the vendors. That is a contract matter, not a legal one, but it is what stops a customer's details reaching a training set.
• Reform. The Privacy Act Review proposed removing the small business exemption. Law-firm summaries of the government's September 2023 response report that it agreed in principle, but only after an impact analysis and further consultation with small businesses. The first reform Act, passed in November 2024, did not remove it, and the Attorney-General's Department's privacy page still describes the remaining agreed-in-principle proposals as being worked into draft provisions. Treat the exemption as something that may narrow, and check that page each year.

Voluntary opt-in is the other route. Section 6EA lets an exempt business choose to be treated as an organisation. The OAIC asks for an application form and a link to your privacy policy, and puts your trading name and ABN on a public register. There is no fee. It signals care to customers, and the cost is that you then owe the full set of principles and can be complained about under them.

What the OAIC says about AI products and personal information

The OAIC has published guidance on commercially available AI products. If your business is covered, its main points apply to everyday tools like a chatbot or a meeting summariser. The ones that matter to a small owner:

• The principles apply to any personal information put into an AI system, and to output that contains personal information.
• Typing customer details into a tool counts as a use of the information if it stays under your control, or a disclosure if it goes to an outside party such as the developer.
• Inferred, wrong or generated content about an identifiable person is itself personal information. A chatbot summary that wrongly says a customer "complained twice" is information about that customer.
• Using AI to infer or generate personal information is a collection, and collection needs to be reasonably necessary and fair. Sensitive information generally needs consent, and notice alone does not count as consent.
• Training an AI model on customer data is hard to justify as a use people would reasonably expect. Check whether the vendor's terms let it use your inputs for training.
• Tell people when they are dealing with an AI system rather than a person, for example a website chatbot.
• Do not treat generative output as fact. Record where an AI-derived entry came from, and keep a person checking accuracy.
• As a best-practice recommendation, the OAIC suggests not entering personal information, and especially sensitive information, into publicly available generative AI tools.

The guidance also asks you to do due diligence before adopting a product, and to keep doing it. It says a product should not be used simply because it is available, and it describes a privacy impact assessment as part of a privacy-by-design approach.

A ten-minute coverage check you can file

Run this once a year and whenever you add an AI tool. Answer from your records rather than from memory, and save the filled-in version.

PRIVACY ACT COVERAGE CHECK  (general information only)
Business: [trading name]            Date checked: [date]
1. Turnover, last financial year: $[figure]   Ever above $3,000,000 since 2002? [Y/N]
2. Do we provide health, counselling, care, fitness-health or child care services? [Y/N]
3. Do we sell, swap or give away customer lists or data for any benefit? [Y/N]
4. Do we deliver services under a government contract or subcontract? [Y/N]
5. Are we related to a bigger company (holding or subsidiary)? [Y/N]
6. Do we run a tenancy database or credit reporting business, or report under anti-money-laundering law? [Y/N]
7. Have we opted in (register entry)? [Y/N]
Result: all N = probably exempt, keep notes. Any Y = treat as covered, take advice.
Reviewer: [name]    Next review: [date]

If you answer yes to any line, move to the next section as if you were covered. If you are unsure, the answer is to ask rather than guess. The OAIC suggests seeking advice from your industry association or a lawyer when in doubt.

Filling in the check: a nursery and a bakery side by side

Take two illustrative businesses. A childcare nursery with 38 children, $1.2 million turnover and a parent email list. A neighbourhood bakery with $0.9 million turnover and a loyalty list.

The nursery now has a clear rule. A daily report can be drafted by AI from a template and non-identifying notes ("two children had a water-play session"), but names, photos, health notes and incident details are kept out of the public tools, or go only into a business-plan tool after a vendor check. The related tutorial on children's data in a nursery goes through that setting in detail.

The bakery can use ChatGPT to draft a loyalty email, but it should do so with a [first name] merge field in its email platform rather than by pasting the customer list into a chat window. That habit costs nothing and keeps the list in one place.

Before and after: a staff rule for pasting customer details

Most privacy problems with AI are small, daily and avoidable: someone pastes a spreadsheet to "tidy it up". Compare a rule that fails with one that works.

Before: "Be careful with customer data when using AI."

After: "Use only the business ChatGPT account, never a personal one. Never paste names, phone numbers, addresses, payment details, health information or children's details. Replace them with [Customer A], [Customer B] or an order number. If a task needs real details, ask the owner first. Check every AI draft before it is saved to a customer record."

The second version names the account, lists the data, offers a substitute and says what to check. Staff can follow it without a legal degree. The tutorial on keeping customer data private when your team uses AI has more wording you can adapt.

Swapping names for placeholders before anything goes in

Where someone must work from real notes, strip the identifiers locally first, in a spreadsheet or document, then paste the cleaned version. Here is an illustrative before and after of a note from a care agency.

Original note (do not paste):
"Mrs [surname] in [suburb] fell on Tuesday, bruising on left hip, daughter [name] called 0400 000 000 to ask for a review."

Cleaned note (safe to paste into a business-plan tool):
"Client A fell on Tuesday. Daughter asked for a review. Draft a polite reply offering a call tomorrow and a reminder to log the incident."

Notice what is still in the cleaned note: a fall is health information, even without a name. If the note could be matched to a person by someone with the rota, it can still be personal information. For a covered care agency, the safer course is to keep incident detail out of AI tools until the vendor, the plan and the settings have been checked.

Checking an AI vendor in 20 minutes

The OAIC's due-diligence questions boil down to five things a small owner can ask of any tool.

1. What is it for, and which data goes in? Write the use in one sentence and list the fields. If you cannot, you are not ready to buy.
2. Does the vendor train on my inputs? Look in the plan's data terms, not the marketing page. Business plans usually say no by default, and consumer plans have a switch.
3. Where is the data processed and who else sees it? The OAIC flags that cloud AI can mean disclosure overseas. Look for a sub-processor list and a data-location option.
4. Can I delete it? Some plans keep chats for a period after you remove them. Note the period.
5. Can I turn off features that share data? Look for connected apps, memory and public links.

Our tutorial on what to check in an AI tool's privacy policy and terms shows where each of those answers hides in a vendor's documents. For ChatGPT in particular, remember the two-seat minimum on ChatGPT Business, so a one-person firm either pays for two seats or uses an individual plan with training switched off.

Where covered businesses slip with everyday AI tools

An illustrative case: a veterinary practice that has already confirmed it is covered. Reception turns on an AI note-taker for phone calls so owners do not have to repeat details. Three weeks later the practice finds the tool has been emailing summaries, including owners' addresses and payment discussions, to a personal mailbox on the receptionist's phone. No one chose to disclose anything. The feature was on by default and nobody had read the settings.

What would have caught it: the five-question vendor check above, a test call with fake details before go-live, and a quarterly look at where summaries are stored. The same pattern appears with meeting recorders. Our tutorial on AI meeting note-takers lists the settings to look for.

A second slip is the website chatbot with no sign that it is a bot. The OAIC says people should be told when they are interacting with AI. A single line at the top of the chat window ("You're chatting with an automated assistant; a person can take over on request") deals with that, and the matching tutorial on telling customers you use AI offers wording.

A third slip is the exports. A music school asks AI to "clean up" its student list, including parents' phone numbers and children's ages, and keeps the result in a chat history for months. If your business is covered, that chat history is a store of personal information, so delete conversations you no longer need and keep the cleaned file in your normal systems.

A short AI use note to keep with the coverage check

If you are covered, or choose to behave as if you are, one page is usually enough. Keep it plain and truthful, because a policy that overclaims is worse than none.

AI USE NOTE  -  [business name]
Tools in use: [tool, plan, account owner]
What goes in: [job types and data fields allowed]
Never goes in: [names with health details, payment data, children's data, passwords]
Who checks outputs before they reach a customer or a record: [role]
Training on our data: [off / not used, source: vendor data terms, date checked]
Customer notice: [where we say we use AI, e.g. chat window, privacy policy]
Review: [month each year] or when a tool, plan or turnover changes

Pair it with your privacy policy. If you are not covered, the note is still useful because it makes staff behaviour consistent and answers questions from larger customers.

Working out whether a breach would reach you

If a business is covered, the notifiable data breaches scheme can apply when a breach is likely to cause serious harm. The OAIC's small-business page links to that scheme rather than explaining it, so read the OAIC's breach pages and ask an adviser how it applies to you. The practical point for AI is this: if an AI tool's account is taken over, everything in the chat history is exposed. Use a unique password and two-step sign-in on the account, give one owner control of any shared account, and remove former staff on their last day.

A quick sum shows why history matters. Say a four-person practice has 60 chats a month, and one in six contains a customer detail. Over a year that is about 120 chats holding personal information. Setting a monthly clear-out, or using a tool's temporary chat mode, cuts the number you must worry about without changing how anyone works.

What to do this week

1. Fill in the coverage check and save it with the date.
2. List every AI tool in use, including the ones staff use on their own accounts. Ask them; do not guess.
3. For each tool, note the plan, whether training is off, and what data goes in.
4. Put the staff rule on one page and make the pasted-data rule specific.
5. Put a reminder in the calendar to repeat the check yearly and when turnover nears $3 million.

If you would rather map this with someone who has seen how small teams use AI day to day, that is what my AI implementation consultation covers. For anything on whether the Act applies to you, the OAIC's small business page and a qualified adviser are the right sources, and the broader data classification tutorial is a good companion for the data-sorting step.

Questions small-business owners ask about the Privacy Act and AI

Can I opt in to the Privacy Act even if I am exempt?

Yes. Section 6EA lets an otherwise exempt small business choose to be treated as an organisation under the Act. You complete the OAIC's opt-in form with a link to your privacy policy, and your trading name and ABN go on a public register. You can opt out later by writing to the OAIC, though complaints about the period you were registered can still be made.

Does turnover mean profit?

No. The OAIC counts all income from all sources in a financial year. It leaves out assets held, capital gains and the proceeds of selling capital items. Costs are not deducted, so a business with thin margins can still pass the $3 million line. A business under a full year old should project its income across a full year.

Is a customer's name and email address personal information?

Yes. Information is personal information when it is about an identifiable person, or when the person could reasonably be identified from it. A name with an email address, a phone number, a photo, a purchase history tied to an account and a free-text note about someone all qualify, even if the note is only your opinion of them.

What if an AI tool makes up a fact about a customer?

The OAIC treats inferred, incorrect or generated content as personal information when it relates to an identifiable person. A made-up line in a customer note is therefore information you hold about them. If your business is covered, that raises accuracy duties, so check outputs and keep a person responsible for what is saved to the customer record.

Do I need a privacy policy if I am exempt?

The Act does not require one from an exempt business, but you need one to opt in, and many platforms, payment providers and app stores ask for a link anyway. Writing a short, truthful one is cheap insurance. Make sure it matches what you do, including any AI tool that touches customer details.

Further reads

• How to Stop AI Tools Training on Your Business Data — Switch off training use on the tools your team already has.
• Are Chat-With-PDF Tools Safe for Contracts and Client Files? — Whether uploading client files to document chat tools is safe.
• AI Tools for Lawyers in Australia: A Safe Starting Checklist — A professional-practice checklist that overlaps with this one.
• Is It Safe to Put Customer Data Into ChatGPT? — What happens to customer details once they are pasted into ChatGPT.
• AI Policy for Aged Care: What to Cover and What Staff Must Not Paste — An AI policy outline for care providers handling health information.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OAIC small business guidance and checklist, OAIC pages on health service providers, trading in personal information, opting in, personal information and notifiable data breaches, OAIC guidance on privacy and commercially available AI products, Attorney-General's Department privacy reform page, ChatGPT and Claude pricing pages.