Australian doctors are quietly folding artificial intelligence into their daily work, from tools that transcribe a consultation in real time to chatbots that draft referral letters, and the pace of that adoption is starting to unsettle privacy advocates and patient-safety experts. A report carried by Canberra CityNews this week has put a spotlight on the trend, warning that the tools flowing into consulting rooms are running ahead of the rules meant to govern them.
The concern is not that clinicians are behaving recklessly. It is that a technology capable of recording, summarising and interpreting the most sensitive conversations a person can have is being taken up faster than health regulators, privacy watchdogs and professional colleges can write clear guidance. For patients, the change is often invisible. You might sit down with your GP, describe a mental health struggle or a family history of illness, and never realise that an AI listening tool is capturing every word to generate the clinical note.
How AI crept into the consulting room
The most common entry point has been the so-called AI medical scribe. These products sit on a doctor’s phone or computer, listen to the appointment, and produce a structured summary that the clinician reviews and pastes into the patient record. Vendors pitch them as an antidote to burnout, and the appeal is obvious. Australian GPs routinely describe administrative load as the single biggest drain on their time, and anything that shaves minutes off note-taking is welcomed in a system under acute workforce strain.
Beyond scribes, some clinicians have begun using general-purpose chatbots such as ChatGPT to draft correspondence, summarise research or even sound out a diagnosis. This is where the risk profile changes sharply. Consumer chatbots were never designed as clinical tools, they may retain and reuse the text typed into them, and they can produce confident answers that are simply wrong. When a doctor pastes identifiable patient details into a service hosted overseas, the information can leave the protections of Australian privacy law without anyone signing off on the transfer.
Where the privacy line sits
Health information is treated as sensitive information under the Privacy Act, which carries stricter handling obligations than ordinary personal data. Practices are expected to gain consent, keep records secure, and be transparent about how information is used. An AI tool that records a conversation, stores audio on a third-party server, and feeds transcripts through a model most patients have never heard of tests every one of those principles at once.
The Office of the Australian Information Commissioner has repeatedly cautioned organisations against pouring personal information into commercial AI products without understanding where it goes. The Royal Australian College of General Practitioners has taken a similar line, publishing guidance that urges members to treat AI scribes as clinical tools requiring the same scrutiny as any other, including patient consent, a clear-eyed view of data storage, and a human check on every output. The Australian Health Practitioner Regulation Agency has also signalled that using AI does not dilute a practitioner’s professional responsibility. If an AI-generated note contains an error and a patient is harmed, the accountability still rests with the doctor.
Two ways of reading the risk
Supporters of the technology argue that the productivity dividend is real and that blocking these tools would be a mistake. A well-designed scribe frees a doctor to make eye contact rather than type, and a system that reduces administrative fatigue may lower the clinical errors that fatigue itself causes. On this view, the answer is not prohibition but proper procurement: choose vendors that store data onshore, that offer enterprise-grade security, that let patients opt out, and that keep a human firmly in the loop.
The more cautious camp counters that consent in a busy clinic is often thin. A patient told, almost in passing, that “we use a tool to help with notes” has not necessarily grasped that their words are being recorded and processed by a machine. There is also the accuracy problem. AI systems can hallucinate, inserting a symptom that was never mentioned or dropping one that was, and a rushed clinician skimming a plausible-looking summary may not catch the slip. In a domain where a single wrong detail can steer a diagnosis, the margin for that kind of error is unforgiving.
What it means for Australia
The stakes here are distinctly national. Australia’s health system runs on a spine of shared digital infrastructure, from My Health Record to the electronic prescribing that now underpins most pharmacy visits, and public trust in that infrastructure is hard-won and easily dented. Every fresh data breach in the health sector, and there have been several high-profile ones, chips away at the confidence that lets people speak candidly to their doctor. If patients come to suspect their consultations are being fed into opaque AI systems, some may simply hold back, and a patient who withholds information is a patient at risk.
There is also a geographic dimension. Rural and remote practices, already stretched thin, stand to gain the most from tools that cut paperwork and extend the reach of scarce clinicians. But those same practices often have the least capacity to vet a vendor’s security claims or negotiate a data-handling contract, which means the productivity upside and the privacy exposure may land hardest in exactly the same places. A national framework that leaves smaller practices to fend for themselves would widen, not close, the gap.
What’s next
Pressure is building for clearer rules. Professional colleges are refining their guidance, the information commissioner has flagged AI as a priority, and the Commonwealth’s broader work on AI regulation, including proposals for mandatory guardrails in high-risk settings, will inevitably touch healthcare. The likeliest near-term outcome is not a ban but a tightening: firmer expectations on consent, a preference for tools that keep data in Australia, mandatory human review of AI-generated notes, and procurement standards that practices can lean on rather than having to reinvent for themselves.
For now, the technology is running slightly ahead of the rulebook, and the burden of judgement falls on individual clinicians deciding, appointment by appointment, whether a given tool is safe enough to switch on. That is a lot to ask of a profession already carrying more than its share. The task for regulators is to hand doctors a framework clear enough that the decision stops being a private gamble and becomes a settled standard of care.
Sources: Canberra CityNews.
