AI in Australian general practice: what's deployed and what's hype

The honest 2026 picture

“AI in healthcare” is genuinely transformational — and also one of the most over-claimed technology categories of the decade. The Australian regulatory framework and the real-world deployment pattern are more interesting than the marketing.

What’s actually true:

• The TGA regulates AI-based software-as-medical-device (SaMD) products under a defined framework with risk-proportionate evidence requirements
• The ARTG includes a growing number of TGA-approved AI products with specific clinical indications
• Several AI categories are integrated into routine AU clinical practice — radiology assistance, atrial fibrillation detection in consumer wearables, pathology assist, clinical-documentation transcription
• The AHPRA Medical Board issued 2024 guidance making practitioner responsibility for AI-informed clinical decisions explicit

What’s overstated:

• “AI will replace doctors” — not by current trajectory or regulatory framework
• “ChatGPT for medical advice” — not a TGA-approved medical device; not validated for clinical indications
• “AI diagnoses anything from a photo” — narrow tasks with validation; broad claims don’t survive scrutiny
• General-AI capabilities applied to medical decision-making without human-in-the-loop validation — explicitly cautioned in AHPRA guidance

This page covers the AU regulatory framework, what’s actually deployed, and the realistic medium-term trajectory.

A. The AU regulatory framework — TGA SaMD

Since 2021, the TGA has regulated software as a medical device (SaMD) under the TGA SaMD framework. The framework classifies AI products by risk:

Class	Risk	Example
Class I	Lowest	General health tracking
Class IIa	Low-medium	Decision-support information for non-serious conditions
Class IIb	Medium-high	Decision-support for serious conditions; non-final diagnosis
Class III	High	Diagnostic AI for life-threatening conditions; AI-driven treatment decisions

Higher risk class requires more substantial clinical evidence before ARTG inclusion. Inclusion is the regulatory minimum — it does not mean the product is comparatively superior to alternatives, just that it has met the appropriate evidence threshold for its claimed indication.

The ARTG database is searchable by sponsor or device name. The registered indication may be narrower than the marketing — same issue as physical medical devices.

AHPRA Medical Board guidance (2024). Released formal guidance on AI use in clinical practice. Key principles:

• The registered practitioner is responsible for clinical decisions, whether or not informed by AI
• Practitioners must understand the AI tool’s capabilities and limitations before using it for clinical decisions
• Patients should be informed when AI substantially influences their care
• Records must reflect human clinical reasoning, not just AI output
• Confidentiality and privacy obligations under the Privacy Act + My Health Record apply

B. What’s actually deployed in AU general practice

Radiology AI. Several products on the ARTG for chest X-ray interpretation, mammography assistance, lung-nodule detection on CT, and intracranial-haemorrhage flagging on CT. RANZCR (Royal Australian and New Zealand College of Radiologists) maintains an AI position statement. Most AU radiology practices use AI assistive tools; specialist radiologists retain final responsibility for interpretation.

Retinal photography for diabetic retinopathy. AU general practices increasingly take retinal photographs in-house for diabetes management. AI-based screening systems (some ARTG-listed) flag images for ophthalmologist review based on retinopathy features. Reduces specialist referral burden for clear-normal cases.

Pathology AI. Slide-analysis tools for some indications (Pap test cervical cytology, certain histopathology). RCPA maintains framework guidance.

Atrial fibrillation detection in consumer wearables. Apple Watch ECG and Fitbit irregular-rhythm notification both have TGA inclusion for atrial fibrillation detection. Increasingly common as part of how AF is first identified in AU general practice.

Clinical-documentation transcription. Several products (some ARTG-listed, some not) transcribe consultation audio to structured clinical notes. Saves 5–15 minutes per consult in adopter practices. Privacy considerations under My Health Record framework apply.

Skin-lesion analysis. Several apps on the ARTG for skin-lesion assistance. Limited indications; AU dermatology position remains that clinical examination + dermoscopy by a trained practitioner is the standard, with AI as adjunct.

Sepsis early-warning systems in some hospital electronic medical records.

Triage assistance in some emergency departments and telehealth services.

C. Where the claims run ahead of the evidence

“ChatGPT for medical advice.” Not a TGA-approved medical device. Not validated for clinical indications. Documented to hallucinate plausible-sounding inaccurate information. Cannot maintain audit trails required for clinical-quality records. Increasingly common patient behaviour; AU general practice position is that patients should bring AI-generated questions to GP consultations rather than acting on them autonomously.

“AI outperforming doctors.” Headlines typically rest on narrow-task validation studies — AI vs specialist on a specific image-interpretation task in trial conditions. The narrow finding doesn’t translate to whole-of-encounter performance, which integrates examination, history-taking, context, continuity, and accountability — none of which generalist AI handles.

Direct-to-consumer AI symptom checkers. Vary in quality; few are TGA-approved; some are explicitly excluded by the developer’s own terms of service from clinical decision-making. Reasonable as triage aids; not substitutes for clinical assessment.

Branded “AI-powered” supplement personalisation. Marketing layer; usually not based on validated personalised prescribing in any clinically meaningful sense. Standard supplement evidence applies regardless of personalisation framing.

Autonomous AI prescribing. No TGA-approved pathway. AHPRA framework requires a registered practitioner to make prescribing decisions. Some workflow automation exists (repeat scripts under specific protocols) but is not autonomous AI prescribing.

“AI-detected food sensitivities” or “gut microbiome AI.” Marketing layer over the same underlying tests (IgG panels, microbiome sequencing) whose AU primary-tier evidence is weak. The AI label doesn’t change the underlying evidence assessment.

D. The realistic medium-term trajectory

For AU general practice over the next 3–5 years:

More documentation transcription, fewer hours spent on notes. Probably the single largest near-term productivity gain.

More retinal photography in GP practices. Diabetic retinopathy screening expanding with AI-assist makes systematic screening feasible at general practice level.

More imaging AI in radiology. Continued integration; specialist interpretation retains responsibility.

More consumer-wearable detection feeding into general practice. Atrial fibrillation today; other arrhythmias, sleep disorders, and biomarkers likely next.

More clinical decision-support reminders. Embedded in practice software, reminders about screening eligibility, drug interactions, and Choosing Wisely recommendations.

Specialty-specific AI in dermatology, pathology, ophthalmology — expanding as ARTG inclusions accumulate.

Continued tightening of practitioner responsibility framework — AHPRA guidance is likely to evolve as deployment increases.

What’s unlikely to happen: autonomous AI replacing GP consultations, AI diagnosing and prescribing without practitioner involvement, ChatGPT becoming a TGA-approved medical device.

For patients: AI integration into care is happening, in narrow validated ways, with practitioner accountability. The bigger threats to good care in AU general practice over the next decade are workforce shortages, fragmented funding models, and access in remote communities — not AI overreach.

(MBS / PBS items verified 2026-05-17 via WebSearch — workspace egress to mbsonline.gov.au + pbs.gov.au still blocked; spot-check confirms current.)

What this article is and is not

This is general health information drawn from the TGA SaMD regulatory framework, AHPRA Medical Board guidance on AI in clinical practice, RACGP and AU specialty college positions, and peer-reviewed AI-in-medicine literature. It is not personal medical advice and does not create a doctor–patient relationship. Decisions about specific AI tools in clinical care are made by the treating clinician within the regulatory framework.

For Australian consumer-friendly sources: HealthDirect, My Health Record, Australian Commission on Safety and Quality in Health Care — Digital Health.

---

Sources cited

1. TGA — Software as a Medical Device (SaMD)
2. TGA — ARTG
3. AHPRA — Medical Board guidance on AI
4. RACGP — AI in general practice
5. RANZCR
6. RCPA
7. Digital Health CRC
8. My Health Record
9. ACSQHC — Digital Health
10. HealthDirect
11. Topol EJ — High-performance medicine (Nat Med 2019)
12. McKinney SM et al. — AI for breast cancer screening (Nature 2020)