Can you put client details into ChatGPT? What Australian clinics need to know before 10 December

Somewhere in your clinic, someone has pasted a client's notes into ChatGPT to tidy up a letter.

They weren't being careless. They were being efficient. They probably had no idea it was a problem.

AI in clinics is fine. Pasting client details into a free chatbot isn't. Between the Privacy Act, Ahpra's guidance and a rule change that lands on 10 December, the gap between those two matters more than it used to. Here's what you need to be across.

General information, not legal advice. For your own situation, talk to a lawyer or your professional indemnity insurer.

The short answer

Not into a free or personal chatbot.

The privacy regulator, the OAIC, says that as best practice, organisations shouldn't put personal information into publicly available generative AI tools. Sensitive information least of all. Health information is about as sensitive as it gets.

Paste a client's name, birth date or session notes into a personal AI account and you've handed it to a third party. You don't control where it's stored, who sees it, or whether it trains the next version. The OAIC's guidance on commercially available AI is blunt about the starting point: the Privacy Act applies to all uses of AI involving personal information.

"We're too small for the Privacy Act." You're not.

Most businesses turning over $3 million or less are exempt. Health service providers aren't.

The OAIC's small business guidance lists health service providers as covered regardless of turnover. That includes services for physical, emotional, psychological and mental health.

A two-physio practice is covered. So is a speech pathologist working alone. Assume the Australian Privacy Principles apply to you in full.

What Ahpra expects

Ahpra's guidance, Meeting your professional obligations when using AI in healthcare, comes down to five things.

• You're still accountable. Whatever the AI produces, your judgement applies. A tool being TGA-approved doesn't change that.
• Know your tool. What it's for, where it falls down, what it does with client data, and whether it learns from it.
• Tell your clients. Ahpra says practitioners should inform patients and clients when they use AI, and consider any concerns they raise. The more the tool does, the more you explain.
• Get consent when their data goes in. Ahpra asks for informed consent, ideally noted in the health record. AI scribes should ask for consent before they start. Recording a consultation without consent can be a criminal matter under some state laws.
• Ring your insurer. Ask whether your cover has conditions for AI.

The change on 10 December

This is the one most clinics haven't heard about.

From 10 December 2026, businesses covered by the Privacy Act have to say in their privacy policy when software makes decisions about people. The OAIC released its guidance and a fact sheet on 30 September.

It applies when all three of these are true:

1. You've arranged for a computer program to make a decision, or do a step substantially and directly related to one.
2. The decision could reasonably be expected to significantly affect someone's rights or interests.
3. The program uses personal information about them.

If it applies, your privacy policy has to describe the kinds of personal information used, the decisions the program makes on its own, and the decisions where it does a substantial step.

One thing catches people out. The program doesn't have to be one you built. If you've arranged for a tool to do it, it counts.

So where's the line for a clinic? A receptionist booking the time a client asked for is unlikely to be making a significant decision about them. A system that decides whether someone is eligible for a service, or which practitioner they see, is much closer. The OAIC's flowchart walks through the test. Get a lawyer to read your privacy policy before the date.

How we build AI that keeps health information in

Rules tell you what not to do. The useful question is how to get real work out of AI without breaking them. Four design choices from systems we've built for health businesses.

Give the agent the decision, not the conversation. A mobile allied health team in Sydney uses an agent we built to decide which practitioner sees which client. Client details are private health information, so it never emails, texts or shares a thing. It proposes. The owner decides. The team talks to the client. (Case study.)

An agent that can't send messages can't send the wrong one.

Nothing goes out without a yes. At a multi-disciplinary clinic, contracts and onboarding packs for new practitioners fill straight from the staff register. Nothing sends until the owner approves the wording.

Lock the keys away. The keys that connect your systems are as sensitive as the data behind them. At the same clinic, keys never sit in code or documents, passwords arrive separately from logins, and a separate Wi-Fi network and locked-down computers protect client health data. (Case study.)

Fixed rules for the facts, AI for the judgement. For an NDIS provider, we designed an agent where fixed rules read the compliance data and the AI only decides when to escalate and how to word the message. Every decision is traceable, which matters more now that privacy policies have to describe them. That agent is designed, not yet built. (Case study.)

A seven-point check for your clinic

1. Write a one-line rule for staff. "No client names, birth dates, notes or contact details go into any AI tool that isn't on our approved list."
2. Keep the approved list. One line per tool: what it's for, and what may go into it.
3. Ask every vendor five questions. Where is the data stored? Is it used for training? Who can see it? How long is it kept? Can it be deleted?
4. Tell clients. Update your intake forms to say where AI is used, and ask for consent.
5. Keep a person in the loop for anything clinical, and anything sent to a client.
6. Check your privacy policy against the 10 December rule.
7. Ring your insurer.

Where to start

Your staff are already using AI. The question is whether you know where.

Find out first. Then decide which uses should stop, and which deserve a tool built for health information.

If you'd like help with that, start with an audit. We'll walk through how your clinic runs and give you a short, written list of what's worth handing to an agent first, and how to do it safely.