Dr. Mara Gordon, a family physician in Camden, N.J., writes that she uses an AI scribe app on her personal smartphone during patient visits. She says the app records the conversation, and a large language model converts the transcript into a draft medical note. She says she asks patients for consent first and sets the phone on the desk in the exam room.

Gordon says the recordings are uploaded to a HIPAA-compliant cloud rather than stored on her phone, and that most scribe systems delete the recording and note after a set period. She says the software company is responsible for safeguarding the data, including encryption and access controls.

Gordon also describes errors. She says the scribe once wrote that a patient with metastatic cancer had been misdiagnosed, which did not reflect their conversation. She says scribes sometimes mistake "I do" for "I do not," and she calls that potentially critical. She quotes Dr. Eileen Kim, an assistant professor of internal medicine at the University of California San Diego, who warns that overburdened doctors may not review AI-generated notes adequately, and that errors can spread through the chart.

Gordon says she has found erroneous information in electronic health records often, and that it is hard to tell how much comes from AI and how much from human error. She says she thinks AI scribes might make the problem worse.