This analysis synthesizes 9 sources published the week ending Oct 6, 2026. Editorial analysis by the PhysEmp Editorial Team.
In the span of one week, the HHS Secretary told a Washington audience that AI can give patients a second opinion “much better informed than any doctor in the country,” and Utah let a software system write first-time prescriptions without a physician signing each one. Neither event moved the legal weight of a clinical decision off the physician. That mismatch, with authority drifting toward software while accountability stays put, is now the central employment question in AI in Physician Employment & Clinical Practice.
The two stories ran on separate tracks, one political and one regulatory. Read together, they describe the same job being redrawn: the physician as the person who reviews, explains and answers for a decision a machine helped make.
What Kennedy said, and how organized medicine answered
Speaking at the MAHA Summit, Robert F. Kennedy Jr. said AI could help “free us from medical tyranny,” and he relayed a claim he attributed to OpenAI CEO Sam Altman that physicians would be committing malpractice if they diagnosed or prescribed without consulting AI first, Becker’s Physician Leadership reported. The summit’s sponsors included OpenAI and Anthropic.
Six organizations responded the next day: the AMA, AAFP, AAP, ACOG, ACP and the American College of Surgeons. Their joint statement said “patient safety, physician expertise and the humanity of clinical practice” should guide how the tools are built and used. Medical Economics paired the statement with AMA survey data showing 81% of physicians now use AI professionally, up from 40.5% in 2023, and 85% want a say in how it is adopted in their practice.
That last number matters more than the rhetoric. Adoption is already widespread; what physicians want settled is who decides which tools enter their workflow, and on what terms. Fierce Healthcare placed the exchange alongside a McKinsey estimate that clinical AI could perform 16% to 22% of U.S. outpatient claims, roughly 2 to 3 billion claims a year. Employers reading that figure are more likely to think about capacity than about second opinions.
Utah’s pilot keeps a physician on the hook
The Utah arrangement shows what autonomy looks like in practice. Nolla Health’s app can issue initial acne prescriptions to adult Utah residents, choosing only from eight pre-approved topical options, under a pilot approved by the state’s Office of Artificial Intelligence Policy. Physicians approve every prescription for the first 100 patients. After that, the system sends prescriptions directly to pharmacies and physicians move to scheduled review and sampled audits. Accounts of the later stages differ in detail: Quartz describes monthly audits of at least one in ten prescriptions beyond 500 patients, while Interesting Engineering describes daily and then weekly physician review. Both report a company figure that clinicians agree with the system’s recommendations in more than 96% of cases, a number that has not been independently verified.
The physician role in Utah’s pilot does not disappear after patient 500. It changes shape, from prescriber to auditor of a sample. That is a different job, covering far more patients per physician, and none of this week’s coverage addressed how it would be paid.
An audit role carries its own risk. Writing in KevinMD, economist Matt Hasan argues that a system right 98 times out of 100 trains reviewers into habitual approval, and the danger arrives on case 99. A 96% agreement rate is high enough to make sampled review feel routine, which is exactly the condition Hasan describes. The pilot runs for one year, renewable for two more. Whatever review cadence regulators keep at the end of that window will become the template other states copy.
Patients want the AI opinion, then penalize the doctor who uses it
Kennedy’s framing assumes patients will welcome AI in the exam room. The evidence this week was mixed. eMarketer, citing survey data, reported that 60% of doctors and nurses say patients frequently arrive with AI-generated health information and that 69% of U.S. adults have used AI for a second opinion after an appointment. Yet only 9% trust AI to manage an ongoing condition.
A study in npj Digital Medicine, summarized by News-Medical, tested how 1,030 participants reacted to physicians who disclosed using AI. Those physicians were rated less warm and less competent, and participants were less willing to choose them, recommend them or follow their advice. The penalty was strongest among participants under 45.
Put those findings side by side and the physician’s position gets awkward. Patients consult chatbots on their own, discount the physician who admits to doing the same, and then expect that physician to reconcile the two. eMarketer’s read is that physicians should expect to spend more of each visit explaining where the chatbot and the clinical guidance part ways, and that time comes out of an appointment slot already built around other work. Looking back at our recent coverage, the patient side of this shift, people arriving with chatbot second opinions in hand, got less attention here than it warranted.
Liability has not moved with authority
If AI is “better informed,” a reasonable physician would expect some of the risk to follow it. The contracts say otherwise. Health care attorney Tatiana Melnik told Medical Economics that malpractice carriers may deny coverage when a physician failed to review AI-generated notes, treating it as a technology problem rather than a malpractice one. She also noted that vendor liability caps typically stop at 12 months of fees and can fall to zero for incidents that surface after a contract ends. Patients in California have already filed class actions against practices that used AI scribes without consent.
The Altman claim Kennedy repeated points the standard of care in one direction: failing to consult AI becomes the negligent act. Melnik’s account of insurer behavior points the other way: relying on AI without checking it becomes the uncovered act. Physicians are exposed on both sides.
For physicians weighing offers, the practical exposure sits in a few contract lines: who selects the AI tools, whether the employer’s malpractice policy explicitly covers AI-assisted decisions, and whether audit or review duties for autonomous systems count toward productivity. For recruiters, employers that can answer those questions in writing will have an easier conversation with candidates than those relying on the joint statement’s language about humanity in practice.
The number to watch
Once Utah’s pilot passes 500 patients, sampled audits replace physician sign-off. The figure worth tracking from that point is the error rate regulators find in the audited sample, which will say more than the agreement rate the company reports, along with whether any malpractice carrier writes policy language for the physicians doing that auditing. If neither arrives before other states copy the model, the physician’s new job will be defined by a pilot’s terms rather than by a negotiated one.
Sources
RFK Jr.: AI second opinions are ‘better informed than any doctor’ – Becker’s Physician Leadership
Six physician groups push back on claims that AI is better informed than doctors – Medical Economics
An AI app is now writing first-time acne prescriptions in Utah without a doctor – Quartz
AI and physician judgment: the risk when AI is mostly right – KevinMD
Robert F. Kennedy Jr.’s AI endorsement could leave doctors explaining more to patients – eMarketer
Doctors faced an AI-use penalty after disclosing AI use to patients – News-Medical
The HIPAA and malpractice risks physicians take on with AI tools – Medical Economics