A physician I read about recently opened a clinic visit by talking to the room. “So I’m here today with …” he announced, not to his patient but to the artificial intelligence listening from a phone on the desk. An ambient scribe was recording the encounter, transcribing it and drafting the note. The doctor was narrating for the machine. The awkwardness of this exchange captures a changing dynamic in healthcare: a once-private exchange is now a permanent, searchable, potentially discoverable record, and no one has decided who it belongs to or what it can be used for.
Ambient AI scribes are spreading faster than almost any healthcare tool in recent memory. About 62.6% of the hospitals that run Epic have adopted an ambient tool, according to a recent study published in the American Journal of Managed Care. The early evidence is encouraging: among the 186 clinicians whose burnout was measured, self-reported burnout fell from 51.9% to 38.8% after 30 days on one ambient scribe, in a 2025 uncontrolled before-and-after study published in JAMA Network Open by vendor-affiliated authors, in which the vendor also facilitated data collection. The step no one is preparing for is the next one: these tools are about to become effectively mandatory, required by health systems to standardize billing and wanted by payers and regulators, hunting fraud, as proof.
The questions that matter are not the technical ones we keep asking, how much time the tool saves or whether it transcribes accurately. They are questions of governance: who controls the recording, what protects the people in the room, and what happens once it is no longer optional. I have spent my career at this intersection, as head of clinical innovation at the Johns Hopkins Armstrong Institute for Patient Safety and Quality, as a chief medical officer in digital health, and as a practicing gastroenterologist. My worry is not whether a scribe works; it’s that we are wiring an always-on recorder into the exam room with none of the safeguards every earlier recording technology eventually required.
We do not have to guess at those safeguards. Three older technologies that put an always-on recorder into a high-stakes conversation already map them: aviation’s cockpit recorders show how to protect the recorded party, courtroom stenography, how to protect what people will say, and meaningful use, what happens when adoption is forced.
Protect the recorded party. When cockpit voice recorders were introduced, pilots resisted, because a continuous recording could be turned against the crew. The black box became workable not because the engineering improved but because the rules did: under international amendments that took effect in 2019, recorder data “may be used only for safety-related purposes with appropriate safeguards, and for criminal proceedings.” The uses are enumerated, and everyday employment and commercial uses are not among them. It stayed an instrument of safety because it was walled off from blame. A physician with an always-on scribe has almost none of those walls; nothing stops a transcript from being pulled into a billing audit, a malpractice case, or a productivity review.
The field’s main answer so far is avoidance: malpractice carrier ProAssurance warns that a retained recording “will undoubtedly be discoverable in litigation” and may contradict the signed note, and tells practices to find out whether their vendor stores the audio at all; health-law counsel are advised to “assume AI-generated documentation may be scrutinized in malpractice claims, privacy actions, or regulatory investigations,” according to a 2026 article published in ABA Health Law. That is risk management by not creating the record, less a safeguard than an admission there are none. Medicine needs the aviation answer instead: enforceable limits on what the recording may be used for, who may access it, and what it can never become. Protections first, deployment second.
Protect what patients will say. Patients are exposed differently: their risk is not liability but candor. A visit works because people say what they would tell no one else, and that depends on a privacy that thins the moment the conversation streams to the cloud. An always-on recording changes what people disclose; the stigmatized symptom, the skipped medication, the drinking downplayed are the first to vanish when the room is plainly listening, and most adults already withhold medically relevant information even without one, according to a 2018 article published in JAMA Network Open. The clinician then treats a sanitized version of the patient, misses the warning sign that was never spoken aloud, and signs a note that looks complete but misleads whoever reads it next. Surveillance in the name of completeness produces the opposite.
Other fields draw the line on purpose, and write it into law. Federal law requires that proceedings in open court be recorded verbatim. In civil cases the parties may step outside that record, but only together and only with the court’s leave, “unless the parties with the approval of the judge shall agree specifically to the contrary;” in criminal cases the statute allows no such exception at all.
And federal law further says, “no transcripts of the proceedings of the court shall be considered as official except those made from the records certified by the reporter or other individual designated to produce the record.” Going off the record is not an informality. It is a bounded, mutual decision, announced and approved, about what the official record will contain, and everyone in the room knows at every moment which side of the line they are on. Medicine has no equivalent for ambient capture. The only recourse today is a physician reaching over to pause the scribe, an improvisation that proves the framework is missing.
Don’t force adoption. The freshest precedent is a cautionary one. Meaningful use pushed electronic health records on physicians under the 2009 HITECH Act and docked Medicare payments from holdouts: “The Medicare payment adjustments began on Jan. 1, 2015, for [eligible professionals].” The deadline-and-penalty structure rushed a billing-oriented, half-built product into universal use with little pressure to improve it; we got note bloat, copy forward errors, and a documented rise in burnout, because adoption was coerced, not earned.
Ambient scribes invite the same coercion by a different lever, billing and audit economics rather than a federal deadline. Payers and regulators already treat documentation as enforcement: federal compliance guidance now tells Medicare Advantage plans to educate staff on “the appropriate use of queries and other prompts” and to review “any software or systems used including those created by vendors for plan use and also those used by providers,” Medicare is shifting to AI-driven fraud detection, and insurers are already responding with automated downcoding, according to an article published by HIT Consultant. Make the always-on transcript the record a clinician is paid and audited against, and the predictable follows: physicians narrating defensively for the microphone, documentation becoming a compliance artifact, and the tool meant to relieve burnout creating the strain of being continuously watched.
None of this is an argument against ambient scribes; used well, they give clinicians back time and attention, the scarcest things in any visit. It is an argument about the terms. We can let vendors and payers set them under financial pressure and watch physicians and patients react the way aircrews and EHR-era doctors already did. Or we can do the harder thing first: protect the recorded party, protect what patients will say, and refuse to make the tool mandatory before it has earned its place, deciding as a profession what an ambient scribe should capture before someone else does.
The black box took decades to earn its trust. Medicine can borrow it now, while the recording can still be turned off.
Dr. Mathews is a practicing gastroenterologist and chair of the American Gastroenterological Association’s Committee for GI Innovation & Technology. He was previously head of clinical innovation at the Johns Hopkins Armstrong Institute for Patient Safety and Quality and a chief medical officer in digital health. Views are his own.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
