Ambient AI scribes, tools that listen to a clinical visit and draft the note, have moved from pilot projects to everyday practice faster than almost any health technology before them. Physicians like them. Many say the tools give back hours each week and let them look at patients instead of keyboards.
This year, the legal questions caught up.
For attorneys, the scribe changes what a note is. It is no longer only what the clinician wrote. It is what a tool heard, what it generated, what the clinician reviewed, and what the clinician signed. For healthcare organizations, it changes what a documentation policy has to cover. Consent, verification, and error tracking all need answers that did not exist two years ago.
What happened in 2026
Consent lawsuits. In April 2026, patients filed a proposed class action in the U.S. District Court for the Northern District of California against Sutter Health and MemorialCare, alleging that ambient AI tools recorded visits without appropriate patient knowledge or consent. A separate suit against Sharp HealthCare makes similar allegations and adds one that should get every documentation reviewer's attention: the patient alleges that the note itself stated consent had been obtained when it had not. California requires all parties to consent to the recording of confidential communications.
Accuracy findings. In May 2026, Ontario's Auditor General reported that all 20 AI scribe products approved for use in that province showed inaccuracies during procurement testing. A 2026 survey of emergency physicians using an ambient scribe found that fewer than half trusted it to document accurately.
Regulation. Most ambient scribes are treated as administrative tools, not medical devices, and so are generally outside FDA oversight. Federal policy has leaned toward accelerating AI adoption. HHS issued a request for information on that subject in December 2025 and discussed the responses in June 2026.
The pattern is familiar from other health technologies: adoption first, oversight later, and litigation filling the gap in between.
Four error patterns worth knowing by name
Published testing keeps finding the same kinds of mistakes. They are worth naming, because each one is found differently in the record.
Omission. Something the patient said never reaches the note. These are the hardest errors to find, because there is nothing on the page. The only way to catch one is to compare the note against another source: a recording, a transcript, the clinician's own notes, or a later account of the visit.
Fabrication. The note contains something never said or done, such as an exam finding that was not checked or counseling that did not happen. Fabrication is found by testing the note against what the visit could actually have contained.
Misattribution. Something was said, but by someone else. A family member's history becomes the patient's. In a visit with more than one voice in the room, this is the error to look for first.
False certainty. An uncertain conversation becomes a confident statement, and the clinician's documented doubt disappears. The note reads as though a decision was made when the visit only raised a question.
These errors fail the test of Record Integrity, the first pillar of The Lexcura Clinical Intelligence Model. A note that omits, fabricates, misattributes, or overstates is not a reliable record of the visit, and every conclusion built on it inherits the problem. The model's pillars, Record Integrity, Baseline, Imaging Analysis, Timeline Reconstruction, Standard of Care, Regulatory and Compliance Overlay, and Causation Mapping, then test what the error changed, if anything.
A case that shows the problem
Consider an illustrative scenario. A 61-year-old woman tells her primary care physician about dizziness and occasional chest pressure when walking uphill. She also mentions anxiety. The AI-drafted note, signed by the physician, focuses on anxiety and states that she "denies chest pain." Five days later she is hospitalized with a heart attack.
The questions:
What was actually said, and is there any recording, transcript, or draft? Ask about retention early. Some systems keep recordings briefly or not at all.
What did the physician type herself, and what did the tool generate? The answer separates the clinician's own documentation from the tool's output, and it is usually found in the audit trail and the draft history.
How long was the note open before it was signed? An audit trail showing a long note opened and signed in twelve seconds says something about the review.
Does the note document consent to recording, and is that accurate? If the consent statement is template text rather than a record of an actual patient response, it proves nothing.
Did the omission change care? If the physician ordered cardiac testing anyway, the bad note and the clinical decision point in different directions. If a later clinician relied on "denies chest pain," the connection is stronger. An inaccurate record and a cause of harm are two separate findings, and both have to be established.
For healthcare organizations
The lawsuits suggest three areas to review now.
Consent workflow. How is consent obtained and documented? Does the note's consent statement come from an actual patient response, or is it template text?
Verification. What does "human review" mean in practice? Can clinicians see which content came from today's conversation and which was pulled from prior records?
Error tracking. Is there a simple way for clinicians to report scribe errors, and does anyone review the reports? A tool whose errors are never reported is not a tool without errors. It is a tool without feedback.
The records that matter
- The signed note and every prior version, with timestamps
- The note's full audit trail
- Any audio, transcript, or draft retained, and the system's retention policy
- The clinician's own notes from the visit
- Consent documentation and the organization's consent policy
- Training materials and policies for the tool
- Product name and version in use
- Every later note or summary that repeated the disputed content
- The clinical record before and after the visit
Frequently Asked Questions
What is an ambient AI scribe?
A tool that listens to a clinical visit and drafts the note. Ambient scribes have moved from pilot projects to everyday practice, and many physicians say they give back hours each week.
What are the four error patterns in AI scribe notes?
Omission, where something the patient said never reaches the note; fabrication, where the note contains something never said or done; misattribution, where something said by someone else is attributed to the patient; and false certainty, where an uncertain conversation becomes a confident statement.
Can a patient sue over being recorded by an AI scribe?
In April 2026, patients filed a proposed class action in the Northern District of California against Sutter Health and MemorialCare, alleging ambient AI tools recorded visits without appropriate patient knowledge or consent. A separate suit against Sharp HealthCare adds an allegation that the note itself stated consent had been obtained when it had not. California requires all parties to consent to the recording of confidential communications.
Does an inaccurate AI-drafted note prove harm?
No. An inaccurate record and a cause of harm are two separate findings. If the physician ordered the right testing anyway, the bad note and the clinical decision point in different directions. If a later clinician relied on the error, the connection is stronger.
Why ask about recording retention early?
Some systems keep recordings briefly or not at all. If the audio, transcript, or draft is gone, the note can only be tested against the surrounding record, and omissions become much harder to prove.
About the Author
Michelle Carroll is the Founder and Chief Clinical Strategist of Lexcura Summit, where she leads clinical intelligence reviews of medical records for attorneys and healthcare organizations.
Talk to Us
When the accuracy of an AI-assisted note is in question, we trace each disputed statement to its source and test whether it changed care. Visit lexcurasummit.com or book a brief conversation at calendly.com/michelle-lexcurasummit.