US eases rules while AI medical notes still make errors

What this is about

On May 14, 2026, KFF Health News reported on a conflict that reaches directly into clinics: AI scribes for medical visits are spreading quickly while the US government is moving to loosen some health IT requirements. This is not about science-fiction diagnosis. It is about something routine: notes, medication lists, electronic health records and whether clinicians can trust automatically generated text.

The report describes Abridge software used at Kaiser Permanente. A psychotherapist in Oakland said the system was “not super useful” for psychological nuance and had to be corrected. At the same time, studies show that these tools can reduce administrative work. That tension is the story: AI can save time, but wrong or unclear notes can influence later clinical decisions.

What AI scribes actually do

AI scribes listen to or process information from a patient visit and turn it into a structured medical note. That can help doctors spend less time on documentation. A JAMA study published in April 2026 across five hospitals found, according to KFF, that heavy users of these products saved more than half an hour of work per day after one year.

The system does not replace clinical responsibility. It can miss nuance, give the wrong weight to phrasing or make important context unclear. In mental health, for example, tone of voice, pace or emotional color can matter more than the exact words. If those details disappear from the record, the next clinician may see a distorted picture.

Why it matters

Health records are not ordinary office text. They are the basis for medication, diagnosis, referrals and billing. KFF quotes Raj Ratwani of MedStar Health warning that there is currently no adequate federal safeguard to vet scribe software. The report also points to proposed rules from the US health IT office that could weaken requirements around user-centered design and AI transparency.

The American Hospital Association wrote in its February 27, 2026 comment that privacy and security criteria, as well as decision-support criteria, should be preserved. That matters because hospitals do want less administrative burden, but they do not treat every safeguard as useless red tape. For real people, the difference is clear: a confusing or wrongly summarized record can lead to bad decisions at the next medical visit.

In plain language

Imagine you are baking bread and someone writes down the recipe for you. If the note says “one tablespoon of salt” instead of “one teaspoon of salt,” the text may look neat, but the bread is ruined. In medicine, the note is the recipe for later care. It has to be not only fast, but correct and understandable.

A practical example

A primary-care practice sees 40 patients in one day. An AI scribe saves an average of two minutes of documentation per visit, or about 80 minutes per day. That is a real gain.

For one patient with new sleep problems and anxiety, however, the system summarizes the visit as “mild restlessness, no acute risks.” The doctor does not fully correct it on a hectic afternoon. Three weeks later, another clinician reads only the record and underestimates the change. This example is fictional, but it shows the core issue: small documentation errors can become large later when records are treated as reliable truth.

Scope and limits

• The KFF report does not prove that all AI scribes are unsafe. It shows a tension between benefit, quality and regulation.
• Time savings are real, but they do not remove the need for human review. Nuance can be especially important in psychiatry, emergency medicine and complex cases.
• The proposed US rules are part of an ongoing regulatory process. The final version may still change.

For clinics, the pragmatic line is clear: use AI scribes only with defined accountability, plan realistic correction time, measure errors systematically and do not remove transparency and usability checks just as the tools are moving faster into daily care.

SEO & GEO keywords

AI healthcare, healthcare AI, AI scribes, Abridge, electronic health records, KFF Health News, JAMA, patient safety, ONC, ASTP, medical documentation, clinical AI