Clinical AI in Australia stopped being a demo. It reads chest x-rays. It drafts consult notes. It sits inside software you already bought. The question for anyone running digital health here is no longer whether to use it. The question is how to use it safely and prove you did.

We build clinical AI, so we will be blunt about where things actually stand.

What is real today

Radiology is the clearest case. I-MED and Harrison.ai built Annalise.ai, and in 2021 I-MED became the first imaging network in the world to roll an AI chest x-ray tool across its clinics. That product flags up to 124 findings on a single chest x-ray. It runs at national scale, every day, across hundreds of radiologists. There is a non-contrast head CT version too. This is production.

Ambient scribes are the other place AI has gone mainstream. They listen to a consult and draft the note. The time savings are real and clinicians like them. Pathology is moving too, with computational tools being trialled in Australian labs for breast and lung cancer work. The Australian Digital Health Agency now frames AI as part of routine workflows under the National Digital Health Strategy 2023 to 2028, and 2025 was the year procurement caught up.

What the rules actually say

Here is the part teams get wrong. The TGA does not regulate AI because it is AI. It regulates based on what the software is for. If your tool influences a clinical decision, it is almost certainly a medical device and needs to be on the ARTG. Software reforms with tighter classification rules came into full effect on 1 November 2024, and they push diagnostic and screening tools into higher risk classes. You can read the TGA's own framing on its software reforms page.

Then there is privacy. The Privacy Act amendments got Royal Assent on 10 December 2024. New transparency duties for automated decisions that significantly affect people start on 10 December 2026. That clock is already running.

The professional bodies have drawn their lines. The RACGP guidance on AI scribes is direct. Get patient consent. Review every note for accuracy. Know where the data is stored. The AMA position from August 2023 is just as firm. A doctor stays responsible for the decision, full stop.

The gap nobody funds

The hard part is not the model. It is everything around it. Consent that patients actually understand. A record of why the AI suggested what it suggested. A human who checked. A way to prove all of that after the fact.

Australia already has a warning here. When it emerged that I-MED had shared patient scans with Harrison.ai to train AI, the question was consent, and the OAIC opened an investigation in 2024. The technology worked. The governance is what got tested. You can read the University of Melbourne write-up on how it happened.

So where does that leave a hospital or a digital health company in 2026? The capability is here and proven. The accountability layer is where projects live or die. Tools that bolt AI onto a workflow will get used for a while. Tools that build consent and oversight into the workflow are the ones that survive an audit.

If you are deciding what to deploy next year, ask yourself one thing. Could you stand behind every AI-assisted decision your service made, in front of a regulator, six months from now?

That is the question we help our clients answer. If you are shaping clinical AI in a hospital or a government program, talk to Coterie Health. We build the AI and the governance that has to come with it.