Your AI scribe is saving you time. Is it also building the case against you?

A physician finishes a visit, glances at the AI-drafted note, clicks approve and moves on to the next patient. Two years later, that note turns up as evidence in a malpractice case. It describes a detail that was never discussed. The software vendor may no longer exist, but the physician’s signature is still on the chart.

That scenario is a hypothetical. Sarah Matt, M.D., MBA, a practicing physician and host of Medical Economics’ “Vital Signs,” laid it out in September. But it captures where the ambient scribe story stands in late 2026. The tools have won over physicians. The legal, billing and contractual questions around them are only now catching up.

“AI tools fail differently than people fail,” said Matt, who is also a member of Medical Economics Editorial Advisory Board. “The output looks correct, and that’s the problem. Errors that look wrong get caught. Errors that look right get signed.”

The tool physicians asked for

Few technologies in medicine have been adopted this quickly, or this willingly. At the University of California, San Francisco (UCSF), roughly 70% of physicians use an AI scribe, and about 90% of them say they love it. Robert Wachter, M.D., chair of UCSF’s department of medicine, explained to Medical Economics in March why the tools have become the first technology physicians actually like.

“We’ve almost gotten to the point where, if we turned it off, we might lose a fair number of doctors,” Wachter said. “It’s almost become an expectation of practice now.”

The burnout data help explain why. At Mass General Brigham, burnout among scribe users fell from 52.6% to 30.7% over 84 days. “There is literally no other intervention in our field that impacts burnout to this extent,” said Rebecca Mishuris, M.D., M.P.H., the system’s chief medical information officer.

The time savings have been more modest than early promises suggested. A JAMA study published in April, covering 8,581 clinicians at five academic health systems, found that adopters saved about 16 minutes of documentation time per eight scheduled patient hours. After-hours EHR time didn’t change significantly. In a flash poll of 103 physicians, presented at an April webinar hosted by Medical Economics and Physicians Practice in conjunction with Heidi Health, 42% said an AI scribe saved them no time on a typical clinic day. Another 20% said it saved more than an hour.

“We thought it would save 5 or 10 minutes a visit. It turns out, it doesn’t really save that much time,” Wachter said. Much of the recovered time goes back into eye contact and conversation with patients, which physicians value more.

That trade-off brings a new risk. Physicians are moving from writing their notes to editing them, and an editor who trusts the draft stops reading closely.

Your signature, your liability

Medical Economics reported in July that, in nearly every scenario, the physician who signs an AI-generated note carries the liability, not the vendor. Legal commentators apply the “captain of the ship” doctrine, which holds the directing physician responsible for errors they could have caught. Under that view, an electronic signature works as an attestation that the physician reviewed the note and vouches for its contents. Vendor contracts typically limit the company’s liability for documentation errors, so the exposure shifts back to the practice. A one-click “note reviewed” attestation is also increasingly seen as weak evidence that a real review happened.

Dan Silverboard, J.D., a health care attorney with Holland & Knight, told Medical Economics in March that the core risk is simple: “Health care providers may simply sign off on whatever recommendation the AI program makes or approve ambient documentation without verifying that it accurately reflects the encounter. Those are the key risks.”

Real-world error rates are low but not trivial. One evaluation found omissions in 18% of physician-reviewed notes and hallucinations in 11.5%. Even so, 94.7% of sampled notes were free of significant errors. Wachter doubts that reviewing every note will catch the rest.

“If I’ve used the scribe in the last 49 notes and they were perfect, am I really going to be fully attentive as I review note No. 50?” he said. “If I’m human, the answer is no.” At UCSF, no technical mechanism confirms that a physician has read a note before signing it.

Matt doesn’t excuse clinicians who never chose the tool. Many frontline clinicians had no say in bringing it into their workflow, she told Medical Economics in August. But whatever they document, “whether it was written by themselves, their resident, their nurse practitioner or by AI, they still signed it, and so it’s still their responsibility,” she said.

Consent is becoming a lawsuit

Consent is where the litigation has already begun. Patients in California have filed proposed class actions against Sharp HealthCare, Sutter Health and MemorialCare. They allege the health systems recorded their visits with an AI scribe without consent, in violation of state privacy law.

Tatiana Melnik, J.D., a Tampa health care attorney, discussed AI vendor contracts and scribe consent with Medical Economics at MGMA’s annual conference, in an interview published in October. She said a sign on the wall isn’t enough.

“My general recommendation to practices is to ask the patient and get their affirmative consent,” Melnik said. Practices also need a plan for patients who say no, she added. She suggested a practical place to document it: “There is no reason why they can’t add AI terminology to their existing notice of privacy practices.”

Some states already require disclosure. Texas Senate Bill 1188 requires practitioners using AI to review AI-generated records and tell patients about the use. Some attorneys read the law to cover any note produced with AI help.

Consent also affects care. A University of Edinburgh review covered by Medical Economics in September found that patients being recorded may hesitate to raise substance use, domestic violence or mental health. The review also found that notes strip out much of the patient’s own narrative. “The experiences of patients are poorly considered, and there are real risks that the patients’ stories are lost,” lead author Lucas Seuren said in a statement.

When the note drives the bill

The billing exposure may be larger than the malpractice exposure. Many scribes now suggest codes as well as notes. Federal regulators have named AI-influenced documentation and coding as a False Claims Act enforcement priority. In 2024, a major hospital system paid $23 million to settle allegations that an automated coding rule assigned higher-paying codes the documentation didn’t support.

“There is no ‘get out of jail free card’ if an AI recommendation results in a higher code than what was actually performed,” Silverboard said.

The add-on code G2211 is especially exposed. Kaif Ruman, a clinical informatics specialist at the ambient documentation company Scribing.io, wrote in September that AI drafts often fall back on generic language such as “conditions stable” or “will continue current regimen.” That language can leave G2211 reimbursement vulnerable in an audit. “To an auditor, these generic statements are nonprobative boilerplate,” Ruman wrote. The note has to show the physician’s reasoning about the longitudinal relationship, not just state that the patient is complex.

Jay Anders, M.D., chief medical officer of Medicomp Systems, made a related point in an August conversation with Medical Economics. He argued that many ambient tools work backward from a billable code instead of building a complete clinical picture. A code that satisfies a payer, he said, isn’t the same as the clinical detail a physician needs to treat the patient.

Read the contract, not the marketing page

The final risk sits in the vendor agreement. Melnik said practices should look past ownership language. Clauses on “usage data,” data retention after termination and de-identification all carry new weight now that re-identification is easier.

She described how a standard damages clause can fail. Many contracts cap vendor liability at 12 months of fees before an incident. “Well, if the incident arises three years after the contract because you’ve allowed them to keep your data post termination, then the damages cap is zero,” she said. Under HIPAA, the practice would then carry most of the risk.

She also expects malpractice insurers to push back. “I also expect medical malpractice carriers to start denying coverage when a medical malpractice issue arises where AI has been used,” she said. A carrier could argue a physician’s unreviewed note is “a technology issue,” not a malpractice issue. The July Medical Economics report advised disclosing AI tools to your carrier before a claim ever comes up.

The audio recording itself is easy to forget. “That recording is protected health information, and it’s not a byproduct of a note,” Matt wrote.

The physician still owns the record

None of this argues against the technology, and none of the sources suggest physicians should give it up. Wachter called scribes the first step toward bigger tools, a way of earning physician trust before AI moves into diagnosis and treatment.

But that trust works in both directions. For now, the scribe writes the note and the physician answers for it, in an exam room, a deposition or a payer audit. Matt reduced the safeguards to four habits that cost almost nothing.

“Tell your patients, read your notes, show your reasoning, and know where the audio actually goes,” she wrote.