Melbourne psychiatrist's AI note-taking rule sharpens healthcare consent questions

A Melbourne psychiatrist is refusing new patients who will not consent to AI note-taking, the Guardian reports. A single clinic’s intake rule has turned into a national argument about consent, medical records and what happens when generative AI reaches the consulting room.

The case matters because it is unfolding in psychiatry. The Guardian said Dr Hemlata Ranga asks new patients to sign a form stating, “I consent for use of AI transcription … to assist in my clinical care.” In this specialty, notes carry trauma histories, diagnoses, medication changes and whatever a patient is prepared to disclose. Someone who hesitates about automation is weighing how their most sensitive history gets captured and stored — a decision that goes well past declining a convenience feature.

AI scribes have already spread across Australian medicine. The RACGP’s guidance on AI scribes treats the tools as part of ordinary general practice, and the Guardian cited college figures showing two in five GPs now use them, roughly double the number a year earlier. An ABC explainer on doctors using AI digital scribes later put national GP use closer to one in four. The estimates differ but point the same way: ambient documentation is becoming routine software, not a niche trial.

The commercial scale is moving just as quickly. Heidi AI says clinicians have run 115 million sessions through its system in the past 18 months, the Guardian reported.

That number says nothing about quality or patient comfort. It does show the policy debate has changed shape. Clinicians are already using AI note-taking — the live dispute is what disclosures, review steps and opt-out protections should surround it once the practice becomes ordinary clinical infrastructure.

Digital Rights Watch has been blunt. Its campaign on AI transcription in healthcare argues patients should be able to refuse these systems without facing repercussions, and the Guardian quoted policy head Tom Sulston calling on regulators and legislators to step up. That concern lands at the centre of the Melbourne case. If the real-world outcome of saying no is finding another psychiatrist, the patient has been given a formal choice with very little leverage. Specialist appointments are already hard to secure. The gap between what the form says and what the patient can actually do is hard to ignore.

Regulators are working from guidance, not a settled national rulebook. The Therapeutic Goods Administration’s page on digital scribes says patients should know when a tool is being used and notes that consent can be withdrawn at any time. It also keeps the clinician, rather than the software, responsible for reviewing the record. That is a useful baseline. It still leaves open harder operational questions: where audio or transcripts are processed, whether a vendor’s infrastructure sits offshore, who inside a practice can access outputs, and how a faulty summary is corrected once it enters a medical file.

Professional guidance is in a similar place. The RACGP material tells doctors to understand a vendor’s privacy settings, to verify outputs and to remain accountable for the note that ends up in the chart. Sensible advice. It also shows how much risk management has been pushed down to individual clinicians and small practices. Software capability can change in months. Professional norms and patient understanding usually move much slower.

Psychiatry makes that mismatch unusually visible. A cardiology appointment revolves around measurements and a treatment plan. A psychiatric session can include abuse histories, suicidal thoughts, family conflict and workplace allegations. Even where a clinician reviews every AI draft before saving it, the knowledge that software is listening can change what a patient says and how they say it. That is why The Conversation’s discussion of whether patients can say no to a doctor using an AI scribe reads as more than an abstract ethics debate. Candour is part of the clinical value being protected.

The governance questions are practical. Patients need to know what is being captured — raw audio, a transcript or only a draft note. They need to know where data is processed and retained, especially if a service depends on offshore cloud infrastructure. They need a plain answer on who is accountable when a summary is incomplete, misleading or attached to the wrong context. Once AI transcription becomes part of ordinary care, those details are what consent actually means.

For digitalblog, the deeper signal is this: Australia’s AI argument is starting to shift. Abstract capability claims are giving way to sector-specific rules for professions that carry confidentiality duties.

Generative AI is leaving office productivity and entering regulated, intimate work. The country has spent years arguing about model capability, copyright and general-purpose safeguards — clinical transcription belongs to a later stage of adoption. The software is now inside the medical record, and unresolved questions about transparency, review and patient choice land inside the doctor-patient relationship.

This Melbourne case is an early stress test for that transition. It does not settle the argument for every AI scribe or every clinician using one. It does show where pressure will build next. If providers want patients to accept AI in the most sensitive parts of care, they will need clearer explanations than a checkbox, firmer answers on data handling and a refusal pathway that is not an exit door from treatment. A profession built on disclosure cannot treat the mechanics of recording as a minor procurement detail. The technology has reached the consulting room. Australian governance now has to work there too.