Australia's federal health department has raised formal concerns about AI scribe technology deployed in GP clinics and specialist practices, warning that the tools operate with minimal oversight, inconsistent patient consent practices, and data-sovereignty risks that clinicians and suppliers may not fully appreciate.

The concerns emerged from Senate estimates briefing documents obtained under FOI laws by Guardian Australia. In a February 2026 briefing, the department noted that AI scribes "have little oversight," and that some suppliers market their products as sitting outside existing regulatory levers or as privacy-compliant — often with limited transparency about how data is processed or stored. A separate April 2026 briefing from the department's AI Expert Advisory Group (AI EAG) acknowledged the productivity and burnout-reduction case for the technology, while flagging that AI scribes share the same quality and accuracy limitations as other large language models, with downstream consequences for patient safety, clinical accountability, and the integrity of national digital health infrastructure.

The department's FOI release, published on 18 June 2026, also identified a specific concern around cloud data routing: some suppliers may be unaware that their platforms transmit patient data outside Australia, which engages Privacy Act obligations and potentially the My Health Records Act — neither of which was designed with ambient audio capture in mind.

A Technology Scaling Faster Than Governance

The speed of adoption is striking. An RACGP online poll found GP use of AI scribes nearly doubled — from 22% in August 2024 to 40% by November 2025. Companies in the sector say the tools have been used hundreds of millions of times globally in the 18 months prior to July 2026. That trajectory is consistent with what the department's AI EAG described as a rapidly evolving market, where the regulatory classification of the tools remains genuinely contested.

Under the Therapeutic Goods Act, AI scribes qualify as medical devices only if they serve a therapeutic purpose. Documentation tools that do not make clinical recommendations can — and routinely do — sit outside the TGA's remit. That classification gap matters: it means there is currently no mandatory pre-market assessment of accuracy, no required adverse-event reporting, and no baseline standard for data handling specific to these products. Australia's health regulator is now actively considering whether safeguards are warranted, but no framework has been announced.

The regulatory picture is not entirely blank. AHPRA's telehealth guidance, updated in October 2025, specifically references AI scribes, advising practitioners to ensure patients receive adequate information and to assess appropriateness before use in virtual consultations. The RACGP advises that GP output from an AI scribe must be carefully verified for accuracy, and that consent must be sought and documented at the start of each consultation. The OAIC has been active too — its 2024–25 digital health annual report addresses AI scribe use across the profession, and its staff met with scribe providers as part of oversight activities that year. In May 2026, the OAIC delivered a keynote on AI scribes at the IAPP Sydney KnowledgeNet event during Privacy Awareness Week. What is missing is binding, AI-specific regulation that unifies these advisory positions into enforceable standards.

Consent and the Self-Censorship Problem

The consent picture is uneven. The department noted significant variation in how clinicians and practices obtained patient consent, and took the position that informed consent requires patients to understand both the benefits and limitations of the specific tool being used — not merely a blanket acknowledgment that a recording is taking place.

The stakes of getting consent wrong are not limited to legal exposure. A Melbourne psychiatrist, reported in May 2026 by Guardian Australia, refused to accept new patients unwilling to consent to AI scribe note-taking, and publicly identified two distinct risks: the possibility of data compromise or leakage, and the likelihood that patients who distrust the technology will self-censor during consultations. The second risk is clinically serious in psychiatry — therapeutic disclosure depends on a patient's confidence in confidentiality. It is less acute but not irrelevant in general practice, where patients may withhold information about substance use, mental health, or sexual health if uncertain how their words are stored and processed.

Dr Elizabeth Deveny, CEO of the Consumer Health Forum, noted that the FOI documents establish that consumers and the government are asking the same questions about AI scribes — an alignment that, in practice, tends to accelerate rather than delay regulatory action.

What Sits on the Horizon

The department's July 2025 Safe and Responsible Artificial Intelligence in Health Care review characterised AI scribes as appearing "very low" risk in some use cases within the AAAiH trust-building roadmap — a framing that, set against the February 2026 Senate estimates briefing, looks like it is being revised in real time. The Victorian Department of Health published its own Ambient AI Scribes Advisory in June 2026. The RACGP updated its AI scribes guidance page as recently as 29 June 2026 and has made a submission to the federal department on AI-related consultation.

Meanwhile, the next frontier is already being mapped. A March 2026 RACGP newsGP piece examined vision-enabled AI scribes capable of processing video as well as audio — a capability that would substantially expand both the clinical utility and the privacy surface of these tools.

The federal department has not announced a consultation timeline or a specific regulatory instrument. What it has done is place on the public record — through FOI — an assessment that the current position is unsustainable. That is how Australian health tech regulation typically begins.