Key Takeaways
- Australian health practices can use AI with patient data when the chosen use meets privacy obligations and professional expectations; the Privacy Act covers health service providers regardless of turnover, according to the Legal Services Commission of South Australia.
- Patient details should stay out of publicly available AI chatbots under the OAIC best-practice guidance summarised by MinterEllison.
- Business AI tools need checks on training settings, storage and processing locations, retention and contract terms before a practice approves patient data use, and Clayton Utz's summary of the OAIC guidance calls for comprehensive due diligence of the vendor.
- GPs must obtain patient consent before using an AI scribe and carefully check the resulting notes, according to the RACGP.
- The RACGP recommends that personal data collected by an AI scribe ideally be stored in Australia, which is guidance for GPs rather than a statutory storage rule for every practice AI tool.
Yes, but only under conditions. I would require a documented review before any tool receives patient details. A business subscription gives you terms to examine; the subscription alone cannot answer whether the proposed use is suitable.
This guide covers patient data specifically. For the broader starting point, see the guide to AI for small business. I checked the sources below on 14 November 2026, and the OAIC guidance is quoted through Clayton Utz's and MinterEllison's summaries; this is general information, not legal or clinical advice.
Does the Privacy Act Apply to a Small Health Practice Using AI?
Yes. The Legal Services Commission of South Australia explains that the Privacy Act applies to all health service providers regardless of turnover. A small GP, dental, physiotherapy or psychology practice cannot rely on low turnover to avoid the Australian Privacy Principles.
Clayton Utz's summary explains the primary-purpose restriction under APP 6. Ask your adviser whether the proposed AI use fits the purpose for which the information was collected or needs another permitted basis.
Is It Safe to Paste Patient Details into ChatGPT, Claude or Copilot?
I would keep patient notes out of personal chatbot accounts. MinterEllison summarises the OAIC's position as best practice not to enter personal information into publicly available generative AI tools.
The exact product and account matter. I checked Anthropic's and Microsoft's own documentation on 14 November 2026:
| Tool or account | What the vendor says | Practice check |
|---|---|---|
| ChatGPT | Business and consumer plans carry different terms; check OpenAI's current documentation directly. | Check current business terms and residency options before considering patient data. |
| Claude consumer and commercial products | Consumer accounts have a model-improvement setting. Commercial inputs and outputs are excluded from training by default. | Confirm the account, settings, retention and contractual protections. |
| Microsoft Copilot with enterprise data protection | Microsoft says prompts, responses and Microsoft Graph data are not used to train foundation models. Controls vary by subscription. | Check the actual plan, enabled agents and web search terms. |
Microsoft's documentation also says agents have their own terms and Bing web search queries receive separate treatment. I would map every route patient information could take. The guide to putting client data into AI tools covers the broader assessment.
What Do Ahpra and the RACGP Expect for AI Scribes and Consent?
Ahpra's guidance centres on accountability, understanding the tool, transparency and informed consent. Practitioners should use that guidance alongside their college's requirements.
The RACGP's AI scribes fact sheet, updated 29 June 2026, says GPs must obtain consent before using a scribe during consultations and carefully check its output for accuracy.
Ask your indemnity insurer how to record consent. RACGP newsGP reporting notes that some medical defence organisations require written consent. Ahpra and the RACGP set these consent expectations for practitioners and scribes; whether the Privacy Act requires consent for a given admin use depends on that use, so ask your adviser.
Does Patient Data Have to Stay in Australia?
The RACGP recommends that personal data collected by an AI scribe ideally be stored in Australia and advises avoiding products where vendors cannot provide that assurance. That is guidance for GPs, rather than a universal statutory storage rule.
I would ask separately where data is processed and stored, then have the proposed arrangement reviewed. An Australian storage answer leaves the processing question open. The guide to whether AI data needs to stay in Australia covers storage questions beyond scribes.
What Checks Should a Practice Run Before AI Touches Patient Data?
I would turn the vendor due diligence and privacy impact assessment approach, together with the RACGP's scribe guidance, into a written approval checklist:
- Define the task. Write down what the tool receives, produces and sends elsewhere.
- Check the contract. Establish training permissions, retention, deletion and security arrangements.
- Map data locations. Ask where processing and storage happen, including connected services.
- Review privacy documentation. Check collection notices, the privacy policy and whether a privacy impact assessment is needed.
- Set the consent process. For consultation recording, establish how staff obtain and record informed consent.
- Assign review and approval. Name the person checking outputs before information enters a patient record.
Budget for vendor assessment, staff instruction and any legal or indemnity review alongside the software. Ask who will do that work and what the review will cost before committing.
I would pause a rollout if a vendor cannot explain storage, terms permit training on patient data, consultation recording has no consent step, or generated notes enter records unchecked.
Where Can AI Help with Practice Admin Versus Clinical Decisions?
I would start with a narrowly defined administrative task: drafting an enquiry reply, preparing an intake summary or drafting a reminder. Patient details in those workflows are still health information, so the same vendor checks and approval steps apply. For examples of the boundary, see AI for allied health clinic admin.
For an Australian practice that wants enquiry response, client intake or invoice reminder workflows built and then run for it, with a person approving anything sensitive, Truespeak is built for exactly that admin work. I spent more than 20 years in commercial roles, and at Nikon I built an AI chatbot that deflected 140+ customer service calls a month.
Truespeak designs, builds and runs automation, including monitoring failures and reviewing exceptions through managed AI operations. Truespeak's work covers enquiry response, client intake and invoice reminders, and it gives no legal advice, so Privacy Act compliance questions belong with your lawyer or indemnity insurer.
Before patient data enters a proposed workflow, take the data map and vendor terms to your legal or indemnity adviser. Use the OAIC's official AI guidance, Ahpra and your college as the reference points.
Frequently Asked Questions
Who is responsible if an AI scribe gets the notes wrong?
RACGP newsGP reporting states that GPs are ultimately responsible for ensuring the patient health record is accurate and up to date. The RACGP's AI scribes fact sheet says the GP must carefully check scribe output for accuracy.
Can a clinic paste patient notes into ChatGPT?
The OAIC best-practice position, as summarised by MinterEllison, is to keep personal information out of publicly available generative AI tools. Assess the exact product, contract and proposed use before approving patient data.
Do we need to update our privacy policy before using AI?
MinterEllison's summary of the OAIC guidance says privacy policies and collection notices should clearly outline when and how AI will access and use personal information. Review those documents before introducing the proposed AI workflow.
Should we do a privacy impact assessment before using AI?
Clayton Utz's summary of the OAIC guidance lists conducting privacy impact assessments among the steps organisations should take. Ask your adviser whether the proposed patient-data workflow needs an assessment before approval.
Does a business AI subscription make patient data use acceptable?
A business subscription alone does not establish suitability. A practice still needs to check training permissions, processing and storage locations, retention, security, contract terms and the proposed workflow.
Sources
Checked 13 Nov 2026.
- The Law Handbook - Legal Services Commission of South Australia
- OAIC clarifies artificial intelligence privacy obligations - MinterEllison
- New OAIC privacy guidance on AI: What you need to know - Clayton Utz
- Artificial intelligence in healthcare - Ahpra
- Artificial intelligence AI scribes - RACGP
- Are AI scribes risking patient safety? - RACGP newsGP
- How do I change my model improvement privacy settings? - Anthropic
- Is my data used for model training? - Anthropic
- Enterprise data protection in Microsoft Copilot - Microsoft
- Guidance on privacy and the use of commercially available AI products - OAIC
