Every ambient AI scribe demo ends the same way. The visit wraps, the note is already written, and you get an evening back.
None show the part at the beginning where you must tell the patient a machine is listening, explain what that means, answer questions, and get their agreement before the visit can begin.
That disclosure is not a ten-second formality. It is a conversation, and in psychiatry, it can derail the opening of the appointment, the minutes where the patient decides how much of themselves to bring into the room. It is the part of the transaction that the vendor doesn’t do. Understanding why matters because the answer determines who is exposed if it goes wrong.
The tool drafts the note. It doesn’t have the conversation
Ambient scribes are sold on time saved, and for many of us, the benefit is real. Charting is the heaviest unpaid tax on clinical work, and a tool that drafts the note from the encounter gives some of that time back. But the pitch leaves out a new obligation at the start of every visit: Disclose that a recording is happening, explain what it does, and obtain the patient’s agreement before the encounter can proceed. The vendor’s business associate agreement designates the company as a data processor. It does not make the company the party who talks to the patient. That conversation belongs to whoever is in the room, which in practice means us.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The legal floor is lower than most of us assume but is not where this stops
Under HIPAA, if the vendor is a properly executed business associate and the recording is used only for treatment, payment, and health care operations, explicit patient authorization is generally not required. State recording-consent law adds another layer: In two-party states like California, Illinois, and Pennsylvania, recording a patient without consent is a wiretapping violation. In one-party states like Texas, a physician’s own consent to the recording satisfies the statute, and the patient’s awareness has, legally, been the physician’s call.
That is changing. As of January 1, 2026, the Texas Responsible AI Governance Act requires physicians to disclose to patients, by the date of service, that an AI system is involved in their care, clearly, conspicuously, in plain language. Senate Bill 1188 already required disclosure for AI used in diagnosis. Texas took what careful clinicians were already doing voluntarily and made it law. Other states are moving in the same direction, so if your jurisdiction hasn’t yet, it is worth assuming it will.
The “consent obtained” checkbox is not consent
Most ambient-scribe workflows have a field that logs that consent happened: a box, a timestamp, an attestation. It looks like protection but is closer to the opposite.
This is not hypothetical. In November 2025, a proposed class action accused a hospital system of using an ambient AI scribe to record more than 100,000 patient encounters in a state that requires every party’s consent to record, and alleged that the resulting notes carried boilerplate language stating patients had been advised of and had consented to the recording, when they had not. The consent field did its job. It said yes but nobody had actually asked.
I have conducted independent medical reviews under Texas statute, applying a defined framework and producing reports built to withstand scrutiny. That work taught me the difference between a review resourced to catch something and a signature that exists only to assign the error to whoever signed. A consent checkbox with nothing behind it is the latter. It records that the disclosure occurred without providing anything that would make it adequate: no language the patient can understand, no account of where the audio goes, nothing the clinician can rely on. The box is checked. The exposure is assigned. The patient’s actual understanding was never resourced.
What does this mean for how you evaluate these tools?
Informed consent is a clinical process, not a legal disclaimer. Clearing the statute does not clear the duty. Increasingly, the statute is catching up to the duty anyway.
Before adopting an ambient scribe, it is worth asking the vendor a question most of us haven’t thought to ask: What does your product give me to say to the patient, and where does the disclosure live once it’s said? If the honest answer is “that part is on you,” you are being handed a liability the product was never built to help you carry. So price that in before deciding whether the time saved is worth it.
This essay is cited in the KevinMD record on the electronic health record.
Erin J. Silvertooth is a dual board-certified psychiatrist and internist based in Austin, Texas, with more than 20 years of clinical practice. She directs TMS and Spravato/esketamine programs, provides neuropsychiatric consultation in brain injury care, and has consulted in sports psychiatry for Olympic, Paralympic, and collegiate athletes for over a decade.
She advises early-stage health technology companies and investors on AI safety and physician liability in digital health. She has served on Texas Medical Board panels evaluating standard of care and as a state-appointed independent medical examiner. She is Harvard-trained in health communication.
Her writing examines AI in psychiatric workflows, physician liability in digital health, interventional psychiatry, and what clinical practice looks like from inside the room. She shares updates on LinkedIn and Substack.
