Doctor Refuses Care After Patient Declines AI Medical Scribe

A physician recently told a patient he would not conduct a medical consultation unless the patient consented to an AI assistant listening to their conversation. The tool in question was Heidi, an AI medical scribe that transcribes doctor-patient discussions and converts them into structured clinical notes.

The incident, first reported by New Atlas, illustrates a tension emerging in healthcare settings as AI transcription tools become standard practice. While these systems promise to reduce administrative burden and let doctors focus on patients rather than paperwork, they're increasingly being presented not as optional conveniences but as prerequisites for care.

The consent problem

Digital Rights Watch has launched a campaign calling for patients to have a legally protected right to refuse AI scribes without losing access to healthcare. According to the organization, people are already being denied care for declining these tools.

Tom Sulston, Head of Policy at Digital Rights Watch, was direct when asked whether consent can be meaningful if refusing affects access to treatment: "No, in short. Consent gathered under duress, such as threatening withdrawal of treatment, does not constitute valid consent."

Meaningful consent requires two elements: patients must have enough information to understand what they're agreeing to, and they must be able to decline without unreasonable consequences. In the case described, the patient had questions about data processing, retention policies, infrastructure providers, and corporate ownership that the physician couldn't clearly answer.

Technical realities versus reassurances

The doctor assured the patient that nothing was "recorded," which Sulston identified as technically impossible. An AI scribe must capture audio, convert it to digital data, and process it before generating a transcript or clinical note. The meaningful distinction is between data that's processed and data that's retained.

Most AI scribe services delete audio recordings after transcription is complete, which sounds like a privacy safeguard. But this creates a verification problem: if the original recording is destroyed, how can anyone later confirm the transcript is accurate? AI systems can mishear words, omit details, or hallucinate information entirely.

The broader pattern

The medical scribe scenario reflects a wider dynamic in modern technology: convenience often moves ahead of comprehension. Streaming services provide instant access while eliminating ownership. Smart devices automate tasks while creating dependencies on remote servers. AI generates outputs without necessarily revealing where information traveled or how conclusions were reached.

Research shows patients are broadly open to medical AI when it assists rather than replaces clinicians, but they place high value on transparency and explainability. The challenge in a consulting room is that patients may be ill, worried, or in pain—not in an ideal position to investigate hosting locations, retention policies, and privacy documentation.

Sulston noted that self-hosting AI tools isn't automatically safer. A clinic running its own system inherits responsibility for security patches, backups, and infrastructure management—tasks a specialized cloud provider might handle better. He suggested "community AI" infrastructure operated for groups of clinics as a potential middle ground.

Why it matters

As AI tools transition from optional enhancements to mandatory infrastructure in healthcare and other sectors, the question of meaningful consent becomes critical. When refusing a technology means losing access to essential services, the choice becomes coercive rather than voluntary. This pattern will likely accelerate as AI systems become embedded in more professional workflows, making it urgent to establish clear standards for patient autonomy and informed consent before these tools become universal.

The details were first reported by New Atlas.