Every clinic owner we speak to has had the same conversation internally this year. Someone on the team has discovered an AI tool, it genuinely saves them half an hour, and nobody is quite sure whether it is allowed. The question usually gets asked as "are we allowed to use AI?" — which is the wrong question, and the reason it rarely gets a useful answer.

The better question for AI automation in an Australian clinic is narrower and much easier to act on: where does the data go, and whose problem is it when something goes wrong?

The rule that decides most of it

Health information is sensitive information under the Privacy Act 1988, which puts it in the most protected category the Act has. That much most practices know.

The part that surprises people sits in Australian Privacy Principle 8, which governs disclosing personal information to an overseas recipient. Before you disclose, you must take reasonable steps to ensure that recipient handles the information in line with the APPs.

And then the sting, in the OAIC's own framing: if the overseas recipient breaches the APPs in handling information you disclosed, your practice is treated as having breached the APPs itself.

Accountability does not transfer with the data. You disclose it, you remain answerable for it — including for what a vendor on another continent does next.

This matters because the overwhelming majority of AI tools a clinic might reach for are operated by vendors based in the United States or Europe. Pasting a patient's history into one of them is not "using a tool". It is a cross-border disclosure of health information, and APP 8 applies to it.

What "reasonable steps" actually means

It is not a vibe. In practice it means being able to answer three questions about any tool that touches patient information:

  1. Where is the data stored and processed? Not where the company is headquartered — where the data physically goes.
  2. What does the data processing agreement say? Specifically, does it bind the vendor to protections equivalent to Australian law, and does it oblige them to notify you of a breach?
  3. Is the patient's express consent on file — if you are relying on consent rather than on the vendor's contractual commitments?

Worth knowing: there is a pathway in the Act for prescribing countries with comparable privacy laws, which would simplify all of this. It is not yet usable — as at late 2026 no country or binding scheme has been approved under it. Anyone telling you a transfer is fine "because the EU has GDPR" is describing a mechanism that does not operate yet.

The December change worth diarising

Reforms to the Act introduce a requirement for privacy policies to disclose the use of automated decision-making that significantly affects individuals, with practices expected to have policies updated by December 2026.

If your clinic uses anything that automatically triages, prioritises, scores or screens patients, that is the clause to read carefully. If you are only automating admin, it likely does not bite — but the policy still has to be accurate about what you do and do not do.

We would confirm the specifics with the OAIC or your indemnity insurer rather than taking any blog's word for the date, this one included.

So what is actually safe to automate?

The useful split is not "AI versus no AI". It is does this touch health information, and does it leave the country.

Comfortably safe — no patient health information involved:

  • Rostering, shift swaps and leave requests
  • Supplier and invoice admin
  • Drafting website copy, blog posts and social captions
  • Internal policy and procedure documents
  • Summarising your own business reporting
  • Answering general enquiries that contain no clinical detail

Needs real thought — patient information involved:

  • Anything touching clinical notes, histories or results
  • Call recordings and transcripts, which routinely contain health information
  • Recall and reminder systems built on patient lists
  • Intake forms, and wherever their contents end up
  • Triage, prioritisation or scoring of any kind

The practical test we give clinics is deliberately blunt: if this tool were breached tomorrow and the contents published, would a patient be identifiable and would they be upset? If yes, the tool needs the three questions above answered before it goes anywhere near your practice.

Why this pushes clinics toward built systems

There is a reason we end up building custom systems for practices rather than pointing them at something off the shelf.

An off-the-shelf AI tool is designed for everyone, which means its data handling is designed for everyone too — usually processed wherever the vendor finds it cheapest, under terms you cannot negotiate. For rostering or a blog draft, that is fine. For anything carrying health information, you inherit a set of decisions you had no part in making, while retaining every bit of the accountability.

A system built for your practice lets you decide what goes where: patient data kept in Australian-hosted infrastructure, AI used only on the parts that carry no identifying clinical information, and a clear record of what happens to what. That is less exciting than "we automated everything", and considerably easier to explain to a regulator.

It is also usually cheaper than people expect, because most of the value sits in the boring half — the booking confirmations, the recalls, the reporting that somebody currently assembles by hand on a Friday afternoon.

Where to start

Do not start with the tool. Start with a list of the twenty things your team does every week that nobody enjoys, and sort them by whether they touch patient information.

Almost always, the top of the no-patient-data column holds several hours a week that can be automated this month with no privacy exposure at all. Begin there. The harder category will still be waiting once you have a policy and a vendor you have actually checked.

None of the above is legal advice, and the Privacy Act is applied to specific circumstances. If you are weighing a particular tool, the OAIC publishes guidance on the APPs, and your indemnity insurer will generally review the question at no cost.

Talk to us about what your clinic could automate