Paste a customer's complaint email into a free chatbot to draft a reply, and you've just sent that customer's name, contact details and grievance to a third party. It takes ten seconds and feels harmless, which is exactly why it happens. AI and privacy sit closer together than most owners realise, and the good news is that the Australian regulator has published clear, readable guidance on how to use AI tools without getting it wrong.
This article pulls that guidance together for Australian small and mid-sized businesses. It's general information, not legal advice. Privacy law depends on your circumstances, so treat the Office of the Australian Information Commissioner (OAIC) as the source of truth and talk to a lawyer if you're unsure where you stand.
AI and privacy: does the Privacy Act cover your business?
Start with the law itself. The OAIC's summary of the Privacy Act 1988 explains that the Act regulates how Australian Government agencies, organisations with an annual turnover of more than $3 million, and some other organisations handle personal information. The obligations themselves are set out in the 13 Australian Privacy Principles, usually shortened to the APPs.
That turnover threshold is the small business exemption, and it's narrower than many owners assume. The OAIC's small business page says most small businesses (turnover of $3 million or less) aren't covered, but some are, regardless of turnover. The list on that page includes:
- health service providers, which the OAIC describes as including medical practitioners, pharmacists, allied health professionals and complementary therapists;
- businesses that trade in personal information, such as selling a customer list;
- contractors providing services under a Commonwealth contract;
- businesses related to a business the Act covers, and those that have chosen to opt in.
That's only part of the list, and the OAIC offers a privacy checklist for small business to work through properly. If you run a clinic of any kind, assume you're covered and check the detail.
My own view, and the one I give clients: even if you are exempt, I'd behave as if you weren't. The exemption is a legal threshold, not a customer expectation. Nobody who hands you their mobile number and home address cares what your turnover was last financial year, and the OAIC itself recommends that small businesses consider opting in.
What the OAIC says about using AI products
In October 2024 the OAIC published guidance on privacy and the use of commercially available AI products. It's aimed at businesses that use AI tools rather than build them, which covers almost everyone reading this. Its headline points, in plain terms:
- Privacy obligations apply to personal information you put into an AI system and to anything it produces that contains personal information.
- Do due diligence before adopting a product: has it been tested for your use, how will people oversee it, what are the privacy and security risks, and who can access what goes in and comes out?
- Update your privacy policy and notices to explain how you use AI, and make sure public-facing AI tools, such as chatbots, are clearly identified as AI.
- If AI generates or infers information about a person, that counts as collecting personal information under APP 3. The guidance says this includes incorrect or made-up information, such as hallucinations, when it's about an identifiable person.
- Under APP 6, personal information should generally be used for the purpose it was collected for, unless the person consents or would reasonably expect the new use and it's related to the original purpose.
The guidance also stresses that due diligence isn't "set and forget". Products change, staff habits drift, and the OAIC expects regular review across the life of the tool.
Free chatbots and business AI tools are not the same thing
One line in the guidance deserves its own section. 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, because of the significant and complex privacy risks involved.
That is the practical dividing line for most small businesses. A free chatbot opened in a browser tab, under a staff member's personal account, is a very different thing from an AI product your business has chosen, contracted and configured. With the first, you usually can't control where the data goes, how long it's kept or whether it's used to improve the provider's models. With the second, you can ask, and get answers in writing.
The guidance makes another point that's easy to miss: once personal information goes into an AI system, it can be very difficult to track or control how it's used, and possibly impossible to remove. That's the argument for setting staff rules now rather than after something has already been pasted in.
Questions to ask an AI vendor
The OAIC's checklists for selecting an AI product suggest several questions, and they translate neatly into a vendor conversation. Before you sign, ask:
- Where is our data hosted and processed? Which countries, and which providers? If personal information leaves Australia, APP 8 on cross-border disclosure comes into play, and the guidance names it specifically.
- Is our data used to train or improve your models, or anyone else's? Get the answer in the contract, not just a sales email, and ask whether you can opt out if the default is yes.
- Who can access what we put in? The OAIC checklist asks directly whether the system's terms let the developer access personal information put into or generated by the system.
- How long is data kept, and how is it deleted? Ask about prompts, transcripts, call recordings and backups, not just the main database.
- What security is in place, and have there been incidents? APP 11 requires reasonable steps to protect personal information, and the checklist suggests looking for any past security incidents.
- How accurate is it for our use, and how do we check it? APP 10 covers the accuracy of personal information, and generative AI can be wrong in ways that look convincing.
A vendor who can't answer these clearly is telling you something. For what it's worth, our own practice at Imagine AI is to tell every client which AI providers process their data and where it's hosted, and that data is never used to train those providers' models. That's the minimum I'd expect from anyone.
What to tell your customers
Transparency is the part businesses most often skip, partly because it feels like paperwork. It's also the part customers notice. Three things are worth doing.
First, update your privacy policy to say, in plain English, that you use AI tools, what for, and whether information is sent to providers overseas. Second, review the notice you give people when you collect their details, such as the fine print on an enquiry form or what your receptionist says. If you're planning to use existing customer data for a new AI purpose that people wouldn't have expected, the OAIC suggests that seeking consent or offering a genuine opt-out is the safer path.
Third, be upfront when people are dealing with AI directly. The OAIC guidance says public-facing tools like chatbots should be clearly identified as AI. We take the same view on phones: the AI receptionists we build always tell callers they're speaking with an AI, and that can't be switched off. Our explainer on what AI receptionists can and can't do goes into the caller experience in more depth.
If your AI tools also send automated emails or texts, privacy isn't the only rulebook. Our plain-English guide to Spam Act rules for automated emails and SMS covers consent and unsubscribe requirements.
Sensitive information needs extra care
Health details, and some other categories the Act treats as sensitive, generally need consent before they're handled. The OAIC guidance warns that many photographs or recordings of people can contain sensitive information, and that consent can't be implied just because someone was told their information would be collected.
For clinics, allied health practices and anyone else who handles health information, that changes the maths. A transcription tool that records consultations, or an AI that summarises patient notes, deserves a proper privacy impact assessment before it goes near a real patient. The OAIC recommends a privacy by design approach, including a privacy impact assessment, for any organisation considering AI products.
Practical steps you can take this month
None of this needs a big project. A sensible first pass looks like this:
- List every AI tool in use. Ask staff directly, because the chatbot someone uses on their phone won't show up on your software bill.
- Set one clear rule. No customer names, contact details or health information goes into free public AI tools. Write it down and tell everyone.
- Check the settings on paid tools. Look for training, data retention and sharing options, and switch off anything you don't need.
- Ask your vendors the six questions above and keep the answers on file.
- Update your privacy policy and collection notices to mention AI.
- Lock down access. Multi-factor sign-in, and staff seeing only the customer data their role needs.
- Diarise a review. Every six months, check the list, the settings and whether anyone has found a new tool.
Keeping customer data in one system, rather than spread across a dozen apps, makes almost every step on that list easier. It's one reason we build AI into the CRM itself, with multi-factor sign-in and role-based access as standard. Our guide to how an AI-first CRM works explains the approach, and our custom CRM development for Australian businesses page covers how a build is scoped.
The rules will keep moving, so make the OAIC's privacy pages your regular reference point and get legal advice for anything specific to your business. And if you'd like a hand working out which of your current tools touch customer data, talk to our team about a 30-minute review call.
— Craig Winder, Founder
