AI is arriving in clinical settings faster than most practices can write a policy for it. An AI scribe transcribes and summarises the consultation, a diagnostic aid flags a finding on an image, a decision support tool suggests a dose or a differential. The appeal is obvious, and so is the risk of assuming the tool now owns part of the decision.
It does not. Under the Health Practitioner Regulation National Law, as in force in each state and territory, the Australian Health Practitioner Regulation Agency (AHPRA) and the National Boards, such as the Medical Board of Australia, regulate registered health practitioners. They regulate your conduct, not the software. When you use AI in care, you remain the accountable person, and the standards you already work under continue to apply.
What the National Law and the National Boards require
The regulatory frame here is not new law written for AI. It is the existing framework applied to a new tool. AHPRA and the National Boards administer the National Law and set the professional standards, codes of conduct, and guidelines that define what safe and ethical practice looks like. Those instruments already cover the situations AI touches.
Practise within your competence
Competence. You are expected to practise within the limits of your training and competence. A tool that produces an output does not extend your scope. If you cannot independently understand and evaluate what an AI system is suggesting, using it does not make the resulting decision defensible.
Protect privacy and health information
Privacy. Patient health information is sensitive, and practitioners must protect it. AI tools that send consultation content to a third party, store recordings, or process data offshore raise real questions about where that information goes and who can access it. The obligation to safeguard patient information does not pause because a vendor is involved.
Obtain appropriate consent
Consent. Consent obligations apply to how you deliver care, and that includes the use of AI. An AI scribe that records a consultation is a clear example: the patient should understand that the consultation is being recorded or processed by a tool, and appropriate consent should be obtained. Consent to treatment is not the same as consent to being recorded by software.
Keep accurate records and do not rely uncritically on output
Records. You are responsible for the accuracy of the clinical record. An AI scribe can draft a note, but the note becomes your record only when you have reviewed it and confirmed it is correct. AI systems can omit, misattribute, or invent detail, so an unchecked AI note is a clinical and professional exposure.
No uncritical reliance. The consistent signal from AHPRA and the National Boards is that existing standards apply and that practitioners must not rely uncritically on AI output. The tool informs your judgment; it does not replace it.
Where AI comes in
The National Law does not single out AI, so the risk shows up through the ordinary duties above. It helps to see the concrete pathways.
AI scribes and note-takers
These record or listen to the consultation and generate a summary or note. The exposures are consent to recording, privacy of the captured content, and the accuracy of the note you sign off. A scribe that quietly runs without the patient's knowledge, or a note adopted without review, converts a convenience into a conduct problem.
Diagnostic aids
Software that flags findings, for example on imaging, can shift attention and anchor a decision. The practitioner remains responsible for the diagnostic conclusion. Deferring to a flag you do not understand, or dismissing a finding the tool missed because you assumed it would catch everything, are both failures of your judgment, not the tool's.
Clinical decision support
Tools that suggest doses, interactions, or management pathways are aids to reasoning. They can be wrong, out of date, or trained on populations unlike your patient. The accountable decision, including the decision to accept or override a suggestion, stays with you.
The line between AHPRA and the TGA
It is worth being precise about who regulates what. The Therapeutic Goods Administration regulates the device or software as a product, including whether it is an appropriate medical device. AHPRA regulates your conduct in using it. A tool can be lawfully supplied and still be used unsafely, and a tool's regulatory status as a product does not discharge your professional obligations as the practitioner using it. Both regimes can apply to the same clinical moment.
What to do
Govern AI use with a written policy. Decide which tools are approved, for what purposes, and under what conditions. Ad hoc adoption by individual practitioners is how consent, privacy, and record gaps appear.
Build consent and privacy in. Where a tool records or processes patient information, ensure appropriate consent is obtained and that the information handling is compliant. Know where consultation data goes and who can access it.
Keep the practitioner as the accountable decision-maker. Treat every AI output as input to your judgment, never a substitute for it. The clinical decision, and the responsibility for it, remain yours.
Verify output against clinical judgment. Review AI-drafted notes before they enter the record, and sense-check diagnostic and decision support suggestions against what you know of the patient. Do not sign what you have not checked.
Match tool use to competence. Only use tools you can meaningfully evaluate, and train the people using them so reliance stays critical rather than automatic.
If you want a fast read on where practitioner AI use sits against these obligations, the free AIRA Health Check is a practical starting point for mapping your consent, privacy, record-keeping, and accountability gaps before they become a conduct question.