Ahpra's AI rules for health practitioners, explained

What Ahpra's guidance says

Ahpra's guidance, Meeting your professional obligations when using AI in healthcare, was last updated on 22 August 2024 and has not been revised in 2026. That is not a gap. The guidance does not create new rules; it explains how the existing codes of conduct for each profession apply when AI is in the room, and those codes have not moved.

• You remain accountable. The practitioner is responsible for safe care and must apply human judgment to every AI output. TGA approval of a tool does not shift that responsibility.
• Understand the tool. Practitioners are expected to know what a tool was trained on, what it is for and where it fails, before they rely on it.
• Be transparent and get consent. Patients should be told when AI is used in their care, and informed consent is required where personal information is entered into a tool, which is every ambient scribe that records a consultation.
• Check the record. AI-generated notes must be reviewed for accuracy before they become part of the clinical record.
• Watch for bias and cover. The guidance also addresses algorithmic bias, the governance employers owe, and whether professional indemnity insurance covers AI use.

The TGA on scribes. Transcribe, or interpret?

The TGA's digital scribes guidance, updated on 30 January 2026, draws a single line that every practice buying a scribe needs to understand. A scribe that only transcribes or summarises what was said is not a medical device. A scribe that analyses or interprets the conversation, for example by generating a diagnosis, a differential or a treatment recommendation the practitioner did not state, is a medical device and must be included in the Australian Register of Therapeutic Goods before it is supplied.

The line moves with software updates. A vendor that adds a “suggested plan” feature to a transcription product has changed its intended purpose, and the practice deploying it should recheck ARTG status after every material update. The TGA has named AI and software-based medical devices as one of its compliance priorities for 2026 to 2027.

Clinical decision support and the 1 November 2026 change

The TGA regulates AI-enabled software as a medical device whenever its intended purpose is diagnosis, prevention, monitoring, prediction, prognosis or treatment. Clinical decision support software that provides diagnostic or treatment recommendations with a clinical purpose is within that definition. A specific exemption exists for some decision support tools, and on 8 September 2026 the TGA amended it to clarify when the exemption applies, with effect from 1 November 2026.

Any developer or practice relying on the exemption should confirm before that date that the product still qualifies. The TGA's broader review of whether therapeutic goods law is fit for AI, reported in July 2025 with fourteen findings, continues through targeted consultations in 2026; no legislative amendment is in force yet. A separate consultation on the Therapeutic Goods Advertising Code, which closed on 3 September 2026, proposed ARTG number display and AI disclosure requirements in software advertising. Its outcome has not been published.

Privacy. Health data in AI tools, and the December deadline

Health information is sensitive information under the Privacy Act, and health service providers of any size are covered regardless of turnover. The OAIC's guidance on AI confirms that the obligations apply to both inputs and outputs. Entering personal information into an AI tool engages the purpose-limitation rules. Information the tool generates or infers about a person counts as a collection. The privacy policy must disclose AI use. The OAIC's stated best practice is blunt: no personal or health information in publicly available chatbots.

From 10 December 2026 a second obligation commences. If a computer program makes, or substantially contributes to, decisions that could significantly affect a person's rights or interests, the privacy policy must list the kinds of decisions and the kinds of personal information used. The OAIC's final guidance of 30 September 2026 names health as a domain where effects are significant and confirms that a clinician signing off does not remove the duty where the program's output was a material input. Triage tools, risk scores and automated recall systems should all be assessed. Our free builder does the drafting.

The clinician's safety guide

The Australian Commission on Safety and Quality in Health Care publishes the AI Clinical Use Guide, the clinician-facing resource the TGA's scribe guidance cross-refers to. It is structured around checks before, while and after you use an AI tool, with companion safety scenarios on ambient scribes and on interpreting medical images. Handing it to clinical staff is the fastest way to turn Ahpra's principles into habits.

What a practice should have in place

• Classify every AI tool by intended purpose against the medical device definition, and confirm ARTG inclusion before deployment where it qualifies.
• Re-assess after every software update. A scribe that starts suggesting diagnoses has changed category.
• Keep practitioners in the loop: human judgment applied to every output, tools tested for fitness before clinical use, and AI-generated notes reviewed before filing.
• Obtain and record informed consent where patient data is entered into a tool, especially consultation-recording scribes. A sign in the waiting room is not consent.
• Do privacy due diligence before adopting any product: where data is stored, whether inputs train the model, which third parties receive it, and no health information in public tools.
• Publish the automated decision-making section of your privacy policy before 10 December 2026 if any system makes or shapes significant decisions about patients.
• Stand up governance: an accountable owner, a risk treatment plan, pre-deployment testing and incident monitoring, including notifiable data breaches. Check that professional indemnity cover extends to AI use.

Those seven items are, in effect, the evidence base for Responsible AI Australia's Embed and Govern tiers. A certified practice can show a patient, an insurer or a regulator that the work has been done and independently reviewed.

Questions people ask