In February 2026, the Therapeutic Goods Administration published updated guidance clarifying how AI-enabled software is regulated as a medical device in Australia. The framework establishes that regulatory oversight depends strictly on a tool’s intended purpose rather than the underlying technology itself, an approach known as technology agnostic regulation.
How the Therapeutic Goods Administration Determines Regulated Software
Not every AI tool used in healthcare faces oversight from federal regulators. According to Phil O’Sullivan, partner and Healthcare Sector Leader at law firm Allens, the critical determination rests on what the software is designed to achieve and the associated risk of that function. If an application aims to diagnose, prevent, monitor, predict, provide a prognosis for, or treat, it generally triggers Australia’s medical device regulations.
Regulated products span multiple clinical specialties. The Therapeutic Goods Administration framework covers mobile phone apps that assist in diagnosing melanoma from photographs, cloud-based analytics capable of predicting patient deterioration, and chatbots that suggest, deliver, or monitor treatments. It also applies to clinical decision support tools using generative AI for recommendations, eye disease screening software for conditions like diabetic retinopathy, and radiology image analysis designed to spot pneumothorax, pneumonia, or tumors.
Conversely, software built purely for administrative functions escapes this classification. O’Sullivan notes that scheduling systems, billing software, drafting clinic letters, note-taking applications, and pure transcription tools remain unregulated when they operate without making clinical recommendations. However, the regulatory boundary remains fluid. The Therapeutic Goods Administration is actively reviewing products marketed as scribes that simultaneously generate diagnostic or treatment suggestions, creating new compliance challenges for practicing physicians.
Evaluating Compliance and Evidence for Clinical Software
The updated February 2026 guidelines extend beyond product classification to detail how existing regulations govern AI-enabled software updates and evidentiary requirements. Manufacturers must now follow rules regarding software modifications that alter a product's performance or intended purpose. While these mandates target product supply rather than individual practitioners, doctors still bear responsibility for evaluating whether a specific tool suits a particular task.
Choosing an appropriate digital tool requires a structured evaluation process before integration into patient care. Clinicians can search the Australian Register of Therapeutic Goods to verify a product’s regulatory status and compare official registrations against vendor promotional claims. While administrative applications may lack register listings, software vendors promoting diagnostic or predictive capabilities without registration warrant direct inquiries regarding clinical validation, training data composition, and patient demographics.
Pro Tip: Healthcare institutions are increasingly establishing internal governance frameworks that dictate approved and prohibited software use cases, particularly regarding patient data. Clinicians working within hospital networks should cross-reference the Australian Register of Therapeutic Goods with their organization’s internal compliance policies.
Maintaining Human Oversight and Professional Liability
Introducing software into medical workflows does not diminish a practitioner’s professional obligations. O’Sullivan emphasizes that a doctor’s duty of care remains active during every patient interaction, advising clinicians to exercise 100% of their clinical judgment and avoid over-relying on recommendations that may contain errors.
The Australian Health Practitioner Regulation Agency reinforces this standard, noting that TGA approval of a tool does not alter a practitioner’s obligation to apply human oversight and judgment. Practitioners must verify that chosen products are fit for purpose and systematically check the accuracy and relevance of records generated by generative AI scribes. This shared responsibility model places the final burden of patient safety squarely on the treating clinician, balancing technological efficiency with rigorous professional accountability.
Frequently Asked Questions About Australian Medical AI Regulations
What triggers Therapeutic Goods Administration oversight for software?
Regulation is triggered by a tool’s intended purpose rather than its underlying technology. Software intended to diagnose, prevent, monitor, predict, provide a prognosis, or treat patients generally falls under the medical device framework.
Are administrative and transcription tools regulated as medical devices?
No. Software dedicated to scheduling, billing, drafting clinic letters, or pure note-taking is generally not regulated, provided it does not generate clinical recommendations.
How can clinicians verify if a software product is officially registered?
Clinicians can search the Australian Register of Therapeutic Goods to check if a product is listed and compare its registered purpose against how the vendor promotes its capabilities.
Keep reading