Yves here. This post disappointingly does not provide a clear answer to the question of whether doctors can force AI medical transcription on to patients. I must confess as a Luddite to being deeply opposed. First, as IM Doc has warned, the error and outright fabrication rate is high, and medical systems make it well-nigh impossible to correct records when bad data gets in (I assume because changes are presumed to represent an intent to defraud insurers). Second, I object to any third party having my biometrics, as in a voice recording.
At a minimum, it is not legal in states that require all-party consent for recording. Those are, per Wikipedia:
California
Connecticut (For electronic recordings only, all parties must be made aware of recordings, with few exceptions.[40] For in-person recordings, the rule is always one party consent.)[41]
Florida
Hawaii (in general a one-party state, but requires two-party consent if the recording device is installed in a private place)
Illinois (listening to, transmitting, or recording non-electronic private conversations require consent by all parties)
Maryland
Massachusetts (only “secret” recordings are banned, but is the only state without a “public location” exception. Despite having a 1968 law imposing general bans on taping wire and oral communications, it was later ruled to violate the First Amendment in the conditions espoused in a case filed by Project Veritas in 2018.[47])
Montana (requires notification only)
New Hampshire
Oregon* (One party for electronic communications, two party for in-person conversations. Law was ruled in 2023 to violate the First Amendment in a case filed by Project Veritas, but this was overturned on appeal, leaving the Oregon law in place.)
Pennsylvania
Washington (however, section three of the Washington law states that permission is given if any of the parties announces that they will be recording the call in a reasonable manner if the recording contains that announcement).
The problem is most people are too well domesticated to ask or make a stink, and are locked in networks or doctor relationships which make them feel they cannot rock the boat.
I did ask one MD who is in a quasi-institutional practice (my strong impression is his hospital does not employ its doctors but lets they operate there on some sort of shared overheads basis) if his hospital used the medical records system Epic. Epic has been strongly pushing for not just audio/AI transcription but even video recordings.
He said yes and I told him of Epic’s AI/recording push. He strongly objected and said he would not subject patients to that. But I doubt most doctors “own” their patients and practice economics to the degree he does.
Moreover, it is not well understood but medical records are not private. The HIPAA protections are limited. They can be disclosed via subpoena. And you can bet that a big AI company and big insurer will not dig in their heels the way a private doctor might.
By Katherine Ruppelt, Nashville Public Radio and Cara Anthony and Blake Farmer, Nashville Public Radio. Cross posted from KFF Health News
Family physician Eric Boose has been using an artificial intelligence tool to get back to what he calls “old-fashioned medicine” — talking with patients face-to-face, without having to type into a computer at the same time.
“I can really just sit there and engage and just focus on them and listen,” said Boose, who practices at Cleveland Clinic.
Roughly two years ago, he started using an AI notetaker app during patient visits. The tool listens while he talks with patients and then automatically generates a visit summary based on the conversation. The summary is usually ready within seconds after the appointment ends.
“It’s taking care of all that tedious work of charting and taking notes during the visit,” he said. “It’s just freeing up a lot more time to get that done, and I can get home to my family earlier.”
Nearly a third of physician practices are using AI scribes and others are working to add the tool, in an effort to cut down on administrative work.
If your practitioner suggests using an AI scribe at your next appointment, here are three things to keep in mind:
1. Clinicians should ask for your permission.
At the start of an appointment, your doctor might ask something like, “Are you OK if I use an AI scribe to help me take notes during this appointment?” A common practice is to accept verbal, not written, consent from patients before turning the tool on. However, the legal requirements for getting permission to record a patient conversation vary by state.
Boose said you can ask to pause the AI scribe at any point, especially to discuss something sensitive. And if you decline altogether, your practitioner will likely return to taking manual notes on a computer.
2. AI scribes make mistakes too, so check their work.
Like other AI tools, medical scribes can “hallucinate,” or spontaneously add errors into a record. AI scribes can also omit important information or miss context clues within a conversation.
Clinicians are supposed to review and edit the AI-generated visit summaries before adding them to a patient’s record. As a patient, it’s a good practice to carefully review your visit summary and contact your health provider if you notice errors.
3. Yes, the AI company could use your data, with limitations.
Companies and health systems that offer AI scribe tools have access to medical data and are subject to federal standards about how they use and store patient data, under the Health Insurance Portability and Accountability Act, more commonly known as HIPAA.
They may use data from your appointment to help improve their software without informing you, said Darius Tahir, who reports on health technology for KFF Health News. “ If information is ‘de-identified,’ which can mean stripping it of identifiers [and] making sure it’s not personally traceable back to people, then it is more free to be used in more ways,” he said. “There are way fewer regulatory requirements.”
If you want to know how your data is being used, ask either your practitioner or medical system for more information. But you might not get a clear answer, Tahir said.
People and Policy
The U.S. health care system will likely continue to integrate AI technology into patient care. The Trump administration strongly supports the development and use of AI, especially in health care. In early 2025, President Donald Trump issued an executive order reducing existing regulations on AI to help the U.S. “retain global leadership of artificial intelligence.” In December, the U.S. Department of Health and Human Services released an AI strategy stating that the department supports “integrating AI to modernize care and public health infrastructure to improve health at the individual and population levels.”
5 comments
Comments are closed.
I wonder at times if doctors realize they are essentially training their replacements (at least in the primary care level). I can’t prove it, but I have a feeling all this data is being fed to train a some beta AI program that can recognize patterns based on patient complaints and the doctor’s responses and recommendations. After absorbing data from thousands if not millions of patient encounters it will be able to give out similar recommendations as a real live doctor.
And eventually we will have an AI lawyer suing an AI doctor with the case presided by an AI jury and AI judge.
I have a feeling you’re right, since anonymized clinical records are being used to train AI “street medicine” programs designed for unlicensed assessment and “care” of unhoused individuals rather than employing more social workers, psychiatric nurses, and street psychiatrists who would provide traditional services. Of course HIPPAA boundaries and ethics verrrrrry unclear, surely this isn’t a Medicaid grift! See a company that’s recently been contracted on LA’s Skid Row: https://www.akidolabs.com/news
I recently had to consult with an occupational therapist for an evaluation here in Quebec, Canada. Although it was not, unfortunately, stated on their website that they use AI for note taking, it was clearly laid out in the consent document that had to be signed before the session. I returned the document signed, but having crossed out the part concerning AI note taking. At the beginning of the session, the therapist discussed the issue with me, and basically to have the notes taken manually I had to pay an extra 100$ (on top of the 250$ regular fee).
Which is basically a way of making medicine a more formal two tier system. Good, human mediated “health care” for those with disposable income and second rate machine mediated “health care” for those who do not.
Considering how machine systems are programmable, this is a hidden mechanism of population control. Whoever writes the algorithms holds the reins of power.
Stay safe.
Ned Ludd died for our sins. Let us offer up a Novena to Saint Luigi for the repose of Ned Ludd’s soul.
At least at our local clinic here in Palo Alto, CA the doctor’s nurse asks whether we assent and I always say no. Most doctors are ok with this. We did have the one doctor who went on to argue with me that by saying no I was making her spend less time with my kid and more time taking notes. I pointed out that she was welcome to spend as much time as she wanted talking to my kid, we had all afternoon. She refused to back down and said that the data was just between her phone and the computer and it was not going anywhere. Presumably this is what she has been trained to say, I doubt she has personally investigated this, but this claim did not pass the laugh test. If it is going from phone to computer, it is going through the cloud, it is being transcribed on a server somewhere so it is stored there, the transcripts and the audio are surely available on all her computers not just the one in the exam room, and I’m sure the AI vendor is making copies to train their AI to get better at transcription. Pretty sure also that both recording and transcript are making it to their QA team in Kenya or India or wherever.
The whole thing left a very sour taste, that she was so pushy on the AI and that the session started on this adversarial note.
My read is that the CEO/CFO types love AI transcription because:
1. upcoding becomes more automatic, the more things the AI detects they can charge for, the better
2. they can use this as a cudgel to make doctors see more patients per hour