Ahpra and the National Boards expect registered health practitioners using AI to stay accountable for every output, understand the tool well enough to use it safely, tell patients when AI is involved, obtain informed consent before AI records or processes personal data, and meet their existing privacy, bias and insurance obligations. The rules are not new; your code of conduct already covers them.
There is no separate "AI law" for clinicians in Australia. Ahpra's position is that the obligations you already hold apply in full the moment an AI tool touches a consultation. This guide covers what that means in practice, which tools sit inside the TGA's remit, and how to check that the AI you are considering will not put your registration at risk.
What is Ahpra's AI guidance for health practitioners?
Ahpra and the National Boards publish Meeting your professional obligations when using Artificial Intelligence in healthcare, which explains how existing responsibilities in the National Boards' codes of conduct apply when practitioners use AI. It spans the regulated professions, so psychologists, psychiatrists, nurses, GPs and allied health practitioners are covered by the same principles.
Two things are worth understanding. First, it is guidance about your conduct, not approval of any product. Ahpra does not maintain a list of approved AI tools, and using a popular product is not a defence if the output was wrong and you filed it anyway.
Second, Ahpra states the guidance will be updated regularly, and it is supported by case studies on newer generative AI tools. Treat it as a living document and check it before rolling a tool out across a practice.
Which AI tools are regulated by the TGA, and which are not?
This is the distinction that causes the most confusion, and Ahpra addresses it directly.
Some AI used in healthcare is regulated by the Therapeutic Goods Administration. The TGA regulates therapeutic goods that meet the definition of a medical device, which includes AI-enabled software where it has a therapeutic use. Diagnostic imaging analysis and clinical decision support software commonly fall inside this boundary.
Generative AI tools used in clinical practice, such as AI scribing, are usually intended for a general purpose. They have no therapeutic use and do not meet the medical device definition, so as Ahpra notes, they are generally not regulated by the TGA. Check whether a product is registered by asking the vendor or searching the Australian Register of Therapeutic Goods.
The practical consequence matters: for most AI scribes and note-drafting tools, no regulator has reviewed the product for clinical safety before it reached you. The assurance work falls to you and your practice. And where a tool is TGA-approved, Ahpra is explicit that approval does not change your responsibility to apply human oversight and judgment.
What obligations does Ahpra expect you to meet?
Ahpra sets out key principles that map onto obligations already in your code of conduct.
Accountability for the output
Regardless of what technology is used, the practitioner remains responsible for delivering safe and quality care. Ahpra expects you to apply human judgment to any AI output, and to test tools to confirm they are fit for purpose before clinical use. If you use an AI scribing tool, you are responsible for checking the accuracy and relevance of the records it creates.
In practice, that means reading the draft note before it is filed, every time. A note you did not read is still a note you signed.
Understanding the tool
Ahpra expects practitioners to understand enough about a tool to use it safely. At a minimum, review the product information: how the model was trained and tested, on which populations, its intended use, its limitations, and the clinical contexts where it should not be used. Intended use matters most, because it tells you when the output is appropriate to rely on. You also need to know whether your data is used to retrain the model, where it is located, and how it is stored.
Transparency with patients
Tell patients when AI is part of their care, and take their concerns seriously. Ahpra scales the expectation to the situation. If AI improves the accuracy of interpreting a diagnostic image, you are not expected to explain the algorithm. If you are using AI to record a consultation, you need to say considerably more about how it works and how it collects and uses the patient's personal information.
Informed consent before data goes in
Where an AI tool requires input of a patient's personal data, patients need to be involved in the decision. Ahpra advises obtaining informed consent and, ideally, noting the patient's response in the health record. An AI scribing tool built on generative AI will generally require personal data, so it generally requires consent.
Ahpra flags a specific risk: consent matters most for AI that records private conversations, because there may be criminal implications if it is not obtained before recording. Recording laws sit in state and territory legislation and are not uniform, so confirm the position where you practise. Ahpra also expects transcription software to include an explicit consent step before proceeding, which is a fair question to put to any vendor.
The obligations that sit underneath
The guidance also points back to standing duties that AI makes more salient: protecting confidentiality and privacy under privacy and health records legislation; understanding the inherent bias that can exist in data and algorithms, including the effect on Aboriginal and Torres Strait Islander patients and others from diverse backgrounds; complying with TGA and state or territory requirements; working within your practice's governance arrangements for implementing and monitoring AI; and holding professional indemnity insurance that covers your practice. On insurance, Ahpra's advice is to consult your provider if you are unsure whether the AI tools you use are covered. Very few practitioners have made that call. It takes ten minutes.
What does the Privacy Act require when AI touches health information?
Health information is sensitive information under the Privacy Act 1988, attracting the strongest protections in the Australian Privacy Principles.
The OAIC's guidance on privacy and the use of commercially available AI products is the companion document to Ahpra's. Three points matter for clinicians:
- As a matter of best practice, the OAIC recommends organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. A consumer chatbot is not a clinical documentation system.
- Due diligence before adoption is expected: what the product was tested for, where training data came from, security risks, and who can access inputs and outputs. The OAIC frames this as ongoing, not set and forget.
- Privacy policies should explain how AI is used, and consent obligations for sensitive information are not satisfied by notification alone.
For the fuller picture on where the risks concentrate, our guide to AI and data security covers the technical side.
What does this mean for psychologists specifically?
Psychologists have an additional layer to consider. The Psychology Board of Australia's Code of conduct took effect on 1 December 2025, replacing the APS Code of Ethics that the Board had adopted since 2010.
The Board has noted that psychologists should consider whether the introduction of the code amounts to a significant change requiring them to re-obtain consent aligned with it. If you are introducing an AI scribe at the same time, handle both in one conversation and one consent form rather than two.
The standards that apply to mental health progress notes do not soften because a model drafted the first version, and our overview of what clinical notes are sets out the baseline.
How do you choose an AI tool that meets these obligations?
Turn the guidance into a purchasing checklist. Before adopting any clinical AI tool, get clear answers to these:
- Intended use and limitations. What is the tool designed to do, and where does the vendor say it should not be used?
- Data location and retention. Where is patient data stored, for how long, and under whose jurisdiction?
- Training on your data. Is client data used to retrain models? If the answer is anything other than a clear no, get it in writing.
- Consent workflow. Does the tool require explicit consent as a step before recording begins, as Ahpra expects?
- Review before filing. Can a clinician review and edit every AI output before it enters the record, or does anything file automatically?
- Regulatory status. Is the tool on the ARTG, and if so, do you still have oversight processes regardless?
- Insurance. Has your indemnity provider confirmed cover for how you intend to use it?
- Governance. Who in the practice owns monitoring, incident handling and periodic review?
PractaLuma is AI-native practice management software for Australian mental-health practices, built around the review-before-filing principle: AI drafts, the clinician approves, nothing enters the client record unreviewed. See how the clinical documentation and practice features fit together, and our security and data handling page for the storage and retention answers.
Still comparing options? Our comparison of AI tools for therapy notes and practical guide to AI medical scribes cover the field in more detail.
Frequently asked questions
Does Ahpra ban AI scribes? No. Ahpra and the National Boards support the safe use of AI in healthcare and recognise its potential to improve health outcomes and reduce administrative burden. What Ahpra requires is that you remain accountable for the output, understand the tool, be transparent, and obtain informed consent.
Do I need written consent to use an AI scribe? Ahpra advises obtaining informed consent and ideally noting the patient's response in the health record. It does not mandate a specific format, but a documented note or signed form is far easier to defend than a recollection. Check your state or territory's recording laws as well, since Ahpra warns there may be criminal implications for recording without consent.
Is my AI scribe approved by the TGA? Probably not, and that is usually expected. Ahpra notes generative AI tools such as AI scribing are usually general purpose, do not have a therapeutic use, and are therefore not regulated by the TGA. You can check any product against the Australian Register of Therapeutic Goods.
Who is liable if an AI-generated note contains an error? You are. Ahpra is unambiguous that the practitioner remains responsible for safe and quality care regardless of the technology used, and must apply human judgment to AI output. Checking the accuracy and relevance of records created by an AI scribe is your responsibility.
Can I paste client information into ChatGPT to draft a letter? This is the case the OAIC addresses most directly, and its best-practice recommendation is not to enter personal information, particularly sensitive information, into publicly available generative AI tools. Health information is sensitive information. Use a tool built for clinical use with contractual terms covering data handling instead.
One last note: Ahpra states its guidance will be updated regularly, so check the source page before a practice-wide rollout rather than relying on any summary, this one included.