Is Your AI Health Tool a Medical Device? The TGA Question

Key Takeaways

• The TGA regulates software as a medical device, so an AI health tool can be a therapeutic good even though it is only code.
• Whether a tool is a medical device turns on its intended purpose, and classification turns on risk.
• Getting the boundary wrong means supplying an unapproved therapeutic good, which is a foundational defect rather than a compliance detail.
• Intended purpose is read from the whole picture of what you build, claim and market, not from a single disclaimer.
• The classification analysis belongs at product design stage, because it shapes the claims, the roadmap and the fundraising story.

1. 1.In Brief
2. 2.Why Software Is a Device
3. 3.Intended Purpose Decides It
4. 4.Then Classification Turns on Risk
5. 5.The Cost of Getting It Wrong
6. 6.The Question Behind the Question
7. 7.Frequently Asked Questions

Somewhere in Australia right now, a founder is describing their product as a wellness app while their pitch deck describes it as clinical decision support. Those two descriptions belong to different regulatory universes. The TGA regulates software as a medical device, and whether your AI health tool is a medical device turns on its intended purpose. If it is one and you have not engaged with the TGA, you are not early stage. You are supplying an unapproved therapeutic good, and that is a sentence no founder wants to hear for the first time from a regulator.

In Brief

• The TGA regulates software as a medical device. Code alone can be a therapeutic good.
• Whether an AI health tool is a medical device turns on its intended purpose. Classification turns on risk.
• Intended purpose is read from everything you build, claim and market, which means it can shift without anyone deciding it should.
• Getting the boundary wrong means supplying an unapproved therapeutic good, a defect that surfaces in diligence, in disputes and in regulator correspondence.
• Australia has no standalone AI Act, so the device question is decided entirely under existing technology neutral therapeutic goods law.

Why Software Is a Device

The instinct that a medical device is a physical thing dies hard, and it kills companies. The TGA regulates software as a medical device, known as SaMD, which means an algorithm can carry the same regulatory character as an implant or an imaging machine. There is no AI carve out. Australia has no standalone AI Act, and health AI is governed through existing technology neutral law, of which the therapeutic goods framework is one of the oldest and least forgiving examples. The framework does not ask whether your product is innovative. It asks what your product is for.

Intended Purpose Decides It

That phrase, what your product is for, is the legal fulcrum. Whether an AI health tool is a medical device turns on its intended purpose, and intended purpose is not a private fact about your engineering. It is read from the whole picture: what the product does, what you claim it does, how you market it, who you sell it to and what a user would reasonably understand it to be for.

This is why the boundary is so treacherous for AI products in particular. The same underlying model can sit on either side of it depending on framing, features and claims. And AI products drift. A tracking feature becomes an alert. An alert becomes a flag for clinical attention. A summary becomes a recommendation. Each release is sensible on its own, and somewhere in the sequence the intended purpose changed without a single meeting deciding it. The marketing team can move you across the boundary in one campaign. So can a testimonial you chose to amplify.

Then Classification Turns on Risk

Deciding you are a medical device is not the end of the analysis. It is the beginning of the second one, because classification turns on risk, and classification drives everything about your pathway: the evidence expected, the scrutiny applied and the time and cost of getting to market. Two products that both qualify as devices can face entirely different journeys. Founders who assess the boundary question without the classification question have half an answer, and half an answer is not a launch plan.

The Cost of Getting It Wrong

Getting the boundary wrong means supplying an unapproved therapeutic good. That is not a fine print problem. It is a foundational defect in the business, and foundational defects surface at the worst moments. In due diligence, when an acquirer's lawyers ask the device question you never answered. In a dispute, when a user says your tool got it wrong and their lawyers discover the regulatory gap. In the regulator's first letter, which arrives after the product has scaled. Every one of those conversations goes differently for the founder who settled the question before launch, in writing, with advice they can show.

There is an opposite error too. Founders who assume everything health adjacent is a device strip the ambition out of their product to stay clear of a boundary they have not actually located, giving up the clinical value that is usually the point. The boundary is navigable in both directions, but only if you know where it is for your product.

The Question Behind the Question

For AI health founders, the device question is one instance of a larger pattern we work on across the licensed professions. AI products that do what regulated professionals do, in law, in financial advice and in health, all face the same structural moment: a perimeter drawn before the product existed, a regulator that has not yet said yes to your model and a first approval that will belong to whoever designs the path best. The device question is where that moment arrives for health AI.

Frequently Asked Questions

Our product is just a wellness app. Is it regulated by the TGA?

The label wellness does not decide anything. Whether your tool is a medical device turns on its intended purpose, read from what it does, what you claim and how it is marketed. Plenty of products described as wellness apps sit closer to the boundary than their founders believe.

A clinician reviews everything our AI produces. Does that take us outside the framework?

Not by itself. Human involvement changes the analysis but it does not automatically settle the device question, which still turns on the product's intended purpose. What the clinician's involvement actually means for your classification is an assessment worth doing properly.

What happens if we launched without checking?

If the tool is a medical device, you may be supplying an unapproved therapeutic good, and the position compounds as the product scales. The options are better the earlier the question is confronted.

Does a disclaimer saying the tool is not medical advice protect us?

A disclaimer is one input into intended purpose, and a weak one if the rest of the product contradicts it. The analysis looks at the whole picture, and so does the regulator.

When in the product lifecycle should this be assessed?

At design stage, before claims are made publicly, and again whenever the roadmap moves the product towards diagnosis, treatment or clinical decision making. The assessment shapes what you can say and build, so it works best as an input rather than an audit.

Working out which side of the device boundary your product sits on? Contact Astris Law, call (07) 3519 5616 or start with our AI and licensed professions page.

This article is for general information purposes only and does not constitute legal advice and should not be relied on as such. While we take reasonable care to ensure the accuracy of the information provided, we make no representations or warranties as to its completeness, currency or reliability. We accept no liability for any loss or damage arising directly or indirectly from the use of, or reliance on, this website's content. You should always seek professional advice tailored to your specific circumstances before acting on any information in this article. Liability limited by a scheme approved under Professional Standards Legislation.