Sự đồng ý của bệnh nhân đối với người ghi chép AI xung quanh

Ambient AI scribes such as Microsoft's DAX Copilot (from Nuance), Abridge, Suki and Nabla listen to the conversation between clinician and patient, transcribe it, and draft a clinical note that the clinician reviews. Because they capture audio of a private conversation, consent involves three overlapping layers. The first layer is recording law. In the United States, federal wiretap law and most states allow a recording when one party to the conversation consents. Roughly a dozen states, including California, Florida, Illinois, Pennsylvania and Washington, require every party to consent. In those states the clinician's own agreement is not enough. The patient must agree, and so must anyone else in the room, such as a family member or an interpreter. Many health systems simply use all-party consent everywhere. The second layer is health privacy. HIPAA generally allows patient information to be used for treatment and health care operations without a separate written authorization, as long as the vendor signs a business associate agreement. So HIPAA alone often does not require consent. It does govern how the audio and transcripts are stored, used and shared. The third layer is ethics and trust. Even where the law is permissive, most professional guidance and health system policies favor informed, explicit permission. A good disclosure tells patients: what is recorded; that AI drafts the note and the clinician reviews it; how long the audio is kept; whether it is used to train models; and that saying no will not affect their care. A common misconception is that a signed intake form covers everything. A blanket form signed months ago may not reflect what a patient wants for a sensitive visit, so many practices confirm out loud each time. Another misconception is that refusals are rare edge cases. Clinics need a practiced fallback workflow so that saying no is easy and carries no penalty.