Memo to SA Health: Australia does not need another AI scribe

SA Health and a few of its doctors are using $1.9m of public system money to compete with groups like Heidi, which started five years ago, already has millions of users, including in hospitals, and has about $165m in funding. Go figure.

---

A University of Adelaide-led research team, backed in part by SA Health and its Local Health Networks, has just secured $1.9 million over three years from the Medical Research Future Fund to expand AUScribe, an ambient AI scribe being trialled in Soth Australia’s public hospital system.

Set against what is a highly successful and innovative commercial Australian AI scribe landscape, this is about as bad an example of inward-looking locals playing with and wasting taxpayers’ money as you’re going to see in the recent history of the Australian healthcare system.

Unfortunately, it’s just one example of a much broader problem in our healthcare system these days: governments are either not checking or caring if public money goes out the door chasing commercial success that already exists and is being served well by the commercial sector.

Either this or government fails to specify very good local suppliers over global behemoth platforms based on the buying IBM principle.

As far as AI scribes go, Australia has some great ones, big and small. It’s a very healthy and well served sector already.

Heidi Health and Lyrebird Health are both Australian-founded, both already have developed integrations into major hospital electronic medical record systems that can be deployed at scale, both are in the running for NSW Health’s own AI scribe tender, and both are already backed by tens of millions of dollars in private capital.

Heidi alone has raised close to $165 million.

Between them, they’re already close to solving the hardest, most expensive parts of the biggest problem for AI scribes: native write-back into hospital EMRs, clinical safety casework, and the unglamorous compliance engineering that turns a research prototype into something a hospital can actually run at scale.

The case made for AUScribe rests on three claims. Each one is not just weak – they are wrong, and the fact that the researchers don’t seem to get that is a bad reflection on SA Health for letting the project continue, sending another $1.9 million down the drain as far as government is concerned.

The first is that patient data never leaves public hands – a direct accusation that Australian scribes aren’t able to, or don’t, manage their data onshore.

Australian commercial scribes already operate onshore cloud infrastructure specifically to satisfy the same public-hospital data laws AUScribe claims as its point of difference.

There is no sovereignty gap here for a purpose-built local product to close. And the framing itself deserves scrutiny: describing the data as safe because it stays in “public hands” quietly assumes the government is the rightful custodian of a patient’s data.

The federal Sharing by Default agenda is built on the opposite premise – that the data belongs to the patient, not the institution holding it – which makes “we keep it” a strange thing for a government project to be proud of.

The second claim is that the tool is trained on local patients, for local patients.

If SA Health contracted a commercial vendor instead, exactly this would happen, except a lot faster, and likely with far more fidelity and better-established guardrails.

This means the same local fine-tuning, but built on top of a vastly larger base of real clinical use across Australia and internationally that a two-year-old, $1.9 million research build cannot match.

Being locally trained isn’t a reason to build over buy. It’s simply what a competent vendor already does as standard, at a scale a grant-funded prototype has no path to reaching.

The third claim, that public ownership lets clinicians build on top of the tool, is fairly bizarre as well: all the AI scribes in Australia we know have been founded by clinicians and the same clinicians continue to play a major role in developing the product.

All this reflects slightly less on the researchers involved – although maybe if they got out a bit they’d realise that what they are doing has already been done, and done a while ago – than on the levels of government supporting and funding such a misguided project.

The SA Health Department and some of the other government-backed funding bodies involved in this project are duty bound to ask the obvious threshold question before money is committed: does a well-funded, already-integrated Australian commercial product already do this, and if so, why are we funding a competitor to it instead of becoming its customer?

This pattern unfortunately isn’t confined to South Australia.

This masthead has previously reported on Primary Health Networks around the country independently commissioning similar, overlapping digital health tools, duplicating capability that already exists elsewhere, uncoordinated with each other or with federal strategy.

AUScribe is simply the clearest recent example of a pattern running through state health departments, PHNs and federal granting bodies alike: pools of money, distributed independently, each answerable only to its own approval committee in its own region.

It’s the worst kind of “not invented in my backyard” dynamic that seems to be an artefact of our federated and fragmented healthcare technology procurement protocols.

In the age of AI, where collaboration and consolidation are so powerful, it’s not really acceptable anymore.

Compare the SA Health project to the instinct behind NSW Health’s scribe procurement, which deliberately uses the scale of a 15-LHD, 70,000-clinician framework to negotiate real leverage with vendors.

That’s more what coordinated public money looks like. A grant to build a new Australian scribe, when we have so many already competing for the same government’s business, is what its absence looks like.

AUScribe’s own stated ambition makes this much harder to defend, not easier.

The project isn’t pitched as a contained internal tool for “special” SA hospitals – which would be bad enough.

It’s explicitly framed as a model “every Australian health system can adopt,” a direct declaration of intent to compete nationally with Heidi and Lyrebird, and all the other local commercial scribes out there, not just serve one state’s hospitals.

If that’s genuinely the goal, the test is straightforward: is AUScribe entered in the NSW scribe tender, where that exact competition is happening right now? We doubt it, as it’s not really up and running in any meaningful manner yet.

Against a roughly $165 million-funded Heidi and a well-capitalised Lyrebird, a $1.9 million research build isn’t in the contest for the business it says it wants to win.

Australia is not ahead in the global AI race, in health or anywhere else.

Catching up will take every cent of real public investment we have, directed very efficiently, not spent quietly rebuilding what already exists.

Someone, at a federal level, needs the authority and the visibility to ask that question before the money is signed off.

Someone in SA needs to revisit the decision, as embarrassing as that may seem.