Australia’s public hospitals have spent the past two years watching a quiet arms race play out in the consulting room. Ambient artificial intelligence scribes, software that listens to a clinical conversation and drafts the notes a doctor would otherwise type up afterwards, have moved from novelty to near-ubiquity, with commercial products from a swelling field of vendors being trialled across the country. Almost all of them share one feature that makes hospital administrators and privacy officers nervous: they are owned by private companies, and the sensitive recordings they process often travel through systems the health service does not control.
Adelaide is trying a different tack, and it has just been handed the money to prove it can work at scale. The University of Adelaide‘s Australian Institute for Machine Learning, better known as AIML, has been awarded $1.9 million over three years from the Commonwealth’s Medical Research Future Fund to expand the trial of AUScribe, a tool its backers describe as the first ambient AI medical scribe built by an Australian government for its own public health system.
What the grant funds
The funding flows from the MRFF’s National Critical Research Infrastructure scheme, a pool designed to underwrite the kind of foundational capability that individual research projects cannot justify on their own. As Pulse+IT reported, the three-year grant is aimed squarely at scaling AUScribe from a contained trial into infrastructure that other researchers and clinicians can build on.
The distinction that AIML is leaning on is ownership. Where a commercial scribe is a product a health service licenses, AUScribe is being positioned as public infrastructure, developed inside a government-linked research institute and designed for a government health system. That framing matters because the single biggest sticking point for hospitals adopting ambient AI has not been whether the technology works, but who holds the data, where the audio is processed, and whether a clinician can be confident that a patient’s most private disclosures are not being used to train someone else’s commercial model.
Why a government-built scribe is different
The case for building rather than buying rests on a few practical arguments. A publicly owned scribe can be tuned to the terminology and workflows of the local health system rather than a generic template. Its data can be kept onshore and inside the health service’s own governance perimeter. And crucially, the health system is not locked into a single vendor’s pricing or product roadmap, a dependency that has burned plenty of public agencies before.
There is a strong local appetite for exactly this kind of sovereignty argument. It echoes the push from the CSIRO’s outgoing chief, Doug Hilton, who has argued that Australia’s edge lies in building specialised, niche models rather than trying to compete head-on with the giants. A scribe purpose-built for South Australian hospitals is a textbook example of that thesis in action.
The counter-argument is equally real, and worth stating plainly. Commercial vendors have poured hundreds of millions into their products, iterate quickly, and arrive with polished integrations into the electronic medical records that clinicians already use. A government-funded project working on a $1.9 million research grant over three years is not going to match that pace of engineering. Sceptics will reasonably ask whether the public sector can maintain and support a clinical-grade tool over the long haul, or whether AUScribe risks becoming another promising pilot that struggles once the grant money runs out. Building software is one thing; running it as dependable clinical infrastructure for a decade is another entirely.
The doctors’ verdict still matters
Whatever the ownership model, the technology only earns its keep if clinicians trust it and patients accept it. That trust is not a given. Concerns about the privacy and safety of AI scribes have already surfaced among Australian doctors, with questions about consent, accuracy and the medico-legal weight of an AI-generated note that a busy clinician signs off without close review. An error baked into a discharge summary or a referral can follow a patient for years.
A government-built tool does not automatically resolve those worries, but it does change the accountability picture. When the health service owns the scribe, the buck stops with a public body answerable to a minister and to patients, rather than with a private vendor whose terms of service can shift. That is a genuine advantage in a debate that has, so far, been dominated by uncertainty about who is on the hook when something goes wrong.
What it means for Australia
AUScribe lands in the middle of a national conversation about how much of the AI stack Australia should own outright. The country has watched sovereign data-centre deals and local model-building efforts multiply over the past year, driven by a growing unease about depending on overseas platforms for critical systems. Health is arguably the most sensitive corner of that debate, because the data involved is the most personal there is.
If AUScribe succeeds, the template is portable. A scribe proven in South Australian public hospitals could be adapted for other states, or its underlying approach could inform how governments procure AI across the board: build the sensitive core in-house, keep the data onshore, and treat the model as public infrastructure rather than a bought-in black box. That would be a meaningful shift for a public sector that has largely been a buyer, not a builder, of frontier technology. It also puts Adelaide, home to AIML and increasingly a hub for applied machine-learning research, at the centre of a policy experiment with national implications.
The risk runs the other way too. If the expanded trial stumbles, or if clinicians drift back to slicker commercial products, it will hand ammunition to those who argue governments should stick to procurement and leave the engineering to the private sector. The next three years of the trial will be watched closely by every state health department weighing the same build-versus-buy question.
What’s next
With the grant secured, the immediate work is scaling the trial beyond its initial footprint, hardening the tool for wider clinical use, and gathering the evidence on accuracy, safety and clinician acceptance that any health technology needs before it can move from pilot to standard practice. The MRFF’s infrastructure framing suggests the ambition is not just a single product but a reusable capability that other health research can plug into.
The harder test comes after the three years are up. Grant funding builds things; sustained operational budgets keep them alive. Whether AUScribe becomes a permanent fixture of the public health system or a well-regarded experiment that fades will depend on decisions that sit well beyond this particular cheque. For now, Adelaide has bought itself the chance to show that a government can build its own AI, keep it, and run it, in one of the most demanding environments there is.
Sources: Pulse+IT.
