Doctors’ offices were once private. But increasingly, artificial intelligence (AI) scribes (also known as digital scribes) are listening in.

These tools can record and transcribe the conversation between doctor and patient, and draft structured clinical notes. Some also produce referral letters and admin outputs, and even update medical records—but only after clinician review and approval.

Some estimates suggest about 1 in 4 Australian GPs are already using an AI scribe. Major hospitals, including children’s hospitals, are also trialing them.

The pitch is simple: less typing for doctors, more eye contact with the patient. But what about patients’ privacy?

Until recently, the AI scribe market has been largely unregulated. But last month the Therapeutic Goods Administration (TGA)—Australia’s medical device regulator—decided some scribes meet the legal definition of a medical device.

Here’s what this will change, and what patients should know—and ask—about AI scribes in the consult room.

What’s changing

Until now, many AI scribe vendors, from Microsoft to rising Australian startups such as Heidi and Lyrebird—and over 120 other providers—have marketed their tools as “productivity” software.

This means they have avoided the scrutiny of medical devices, which the TGA regulates.

Now, the TGA has found some AI scribes meet the definition of a medical device, especially if they go beyond transcription to suggest diagnoses or treatments.

Medical devices must be registered with the TGA, shown to be safe and do what they claim, and any safety problems or malfunctions must be reported.

The TGA has begun compliance reviews, with penalties for unregistered AI scribes.

This follows similar developments overseas. In June 2025, the United Kingdom health authorities announced tools that transcribe and summarize will be treated as medical devices.

Although still evolving, there are signs the United States will move in a similar direction, and the European Union may too.

In Australia, the TGA has only just begun reviewing AI scribes, so patients can’t assume they’ve been tested to the same standard as other medical products.

What patients should know about AI scribes

They can help—but they are not perfect. Doctors report spending less time on keyboards, and some patients report better conversations.

But tools built on large language models can “hallucinate”—add details never said. One 2024 case study recorded casual remarks about a patient’s hands, feet and mouth as a diagnosis of hand, foot and mouth disease. The potential for errors means clinicians still need to review the note before it enters your record.

Performance varies. Accuracy dips with accents, background noise and jargon. In a health system as multicultural as Australia’s, errors across accents and languages are a safety issue.

The Royal Australian College of General Practitioners warns poorly designed tools can shift hidden work back to clinicians, who then spend extra time correcting notes. Research has found products’ time-saving claims are often overstated once review and correction time is included, underlining the need for devices to be evaluated independently.

Privacy matters. Health data is already a target for hackers and scammers, as the 2022 Medibank breach showed. In recent research with colleagues, we found unsecured third-party applications and lax data protection are among the leading causes of health data breaches.

Clinicians need a clear “pause” option and should avoid use in sensitive consults (for example, discussions about family violence, substance use or legal matters).

Companies must be explicit about where the audio and data are stored, who can access it, and how long it is kept. In practice, policies vary: some store recordings on overseas cloud servers while others keep transcripts short-term and onshore.

A lack of transparency means it’s often unclear whether data can be traced back to individual patients or reused to train AI.

Consent is not a tick box. Clinicians should tell you when recording is on and explain risks and benefits. You should be able to say no without jeopardizing care. One recent case in Australia saw a patient have to cancel a A$1,300 appointment, after they declined a scribe and the clinic refused to proceed.

For Aboriginal and Torres Strait Islander patients, consent should reflect community norms and data sovereignty, especially if notes are used to train AI.

Five practical questions to ask your doctor

1. Is this tool approved? Is it the clinic’s standard practice to use this tool, and does it require TGA registration for this use?
2. Who can access my data? Where is the audio stored, for how long, and is it used to train the system?
3. Can we pause or opt out? Is there a clear pause button and a non-AI alternative for sensitive topics?
4. Do you review the note before it goes into my record? Is the output always treated as a draft until you sign off?
5. What happens if the AI gets it wrong? Is there an audit trail linking the note back to the original audio so errors can be traced and fixed quickly?

Safer care, not just faster notes

Right now, the burden of ensuring AI scribes are used safely rests disproportionately on individual doctors and patients. The TGA’s decision to classify some scribes as medical devices is a positive move, but it is only a first step.

We also need:

• the TGA, professional bodies and researchers to work together on clear standards for consent, data retention and training
• independent evaluations of how these tools perform in real consults
• risk-based rules and stronger enforcement, adapted to AI software rather than traditional devices.

Strong rules also weed out flimsy products: if a tool cannot show it is safe and secure, it should not be in the consult room.

This article is republished from The Conversation under a Creative Commons license. Read the original article.