This analysis synthesizes 7 sources published the week ending Nov 17, 2026. Editorial analysis by the PhysEmp Editorial Team.
Patients have now put a number on what a physician’s review of AI is worth to them. In a West Health-Gallup survey of 5,330 adults who had seen a primary care provider in the past year, 56% said they were comfortable with AI making diagnosis or treatment decisions that a provider could override. Take away the provider’s ability to review the decision, and comfort fell to 9%, Medical Daily reported. The pattern held for imaging (53% versus 13%) and medication decisions (49% versus 11%).
The same week, the AMA House of Delegates adopted policy stating that AI cannot be granted physician licensure, professional titles or certification to practice medicine. Earlier this month we looked at the AMA’s October framework for assigning AI liability. What is new is that delegates voted on Nov. 9 to make those principles formal policy and paired them with a licensure line, including model legislation to keep nonhuman entities from using physician-reserved titles. Read alongside the survey data, this week’s coverage on AI in Physician Employment & Clinical Practice describes a market where the physician’s review is the part patients want most and the part employers have done the least to prepare for.
Physicians underestimate the demand for their own oversight
The Gallup work also surveyed 1,007 primary care physicians. Only 36% thought most patients would accept an AI diagnosis that a provider could override, 20 points below what patients actually said. Physicians underrate how much of that acceptance rests on them.
Use is not the bottleneck. In the same survey, 89% of primary care physicians had used AI for at least one of seven work tasks in the past year, and 66% did so weekly or more. Doximity’s survey of 3,379 physicians in September found 52% using AI for work daily, up from 37% a year earlier. Three in four had run into a factual inaccuracy in the past 12 months.
Preparation is the bottleneck. Only 20% of the primary care physicians in the Gallup survey had formal AI training, and the share falls with practice size: 30% in hospitals and health systems, 16% in physician groups, 11% in independent practices. In Doximity’s data, only 28% of physicians reported clear AI policies and approvals where they work, while 82% said physicians should be the ones making the rules.
The review patients are asking for already happens, informally. Two-thirds of physicians in Doximity’s survey verify AI clinical information all or almost all of the time. That is supervision work being done today, unmeasured, inside schedules that were built before the tools arrived.
What the AMA put in writing
The adopted policy reads like a job description for that supervision. AI-enabled tools should sit inside physician-led teams and be used at the direction of the treating physician, according to the AMA’s announcement. Tools used for diagnosis, prescribing and care management should come with transparent, auditable evidence of safety and efficacy. The AMA will push for a legal framework that protects a physician who declines to use an AI tool or follow its recommendation.
The liability provisions go further. The AMA wants vendors held to the same product-liability standards as other device makers, product liability insurance made a condition of FDA marketing authorization, and disclosure of known failure rates, including across patient populations. It also wants to prohibit end-user license agreements and vendor contracts that shift liability for AI failures onto physicians or health systems. The FDA had authorized more than 1,450 AI-enabled medical devices as of December 2025, the AMA noted.
For employed physicians, one clause stands out. When a hospital, health system or payer mandates a specific AI tool that causes harm, the AMA says that organization and the vendor should bear the liability. Medical Daily pointed out that this could matter for doctors at large systems rolling out AI, and also that the policy is advocacy, not law; courts will keep applying existing malpractice and product-liability law case by case. Until legislatures act, the allocation of risk lives in employment agreements and vendor contracts, which makes “which tools will I be required to use, and who indemnifies me” a fair question in any offer conversation.
AMA President Willie Underwood III, MD, framed the policy around patients: “patients deserve to know who is responsible for their care.” The Gallup data suggests patients have already answered that for themselves.
Who absorbs the cost of reviewing
Physicians are mostly calm about their jobs. In Doximity’s 2026 compensation survey, summarized by The White Coat Investor, 78% said AI had increased their job security or left it unchanged. But 20% already face higher productivity expectations because of AI. Asked who should benefit from efficiency gains, 44% said physicians should make more money, 17% said patients through lower costs, 20% favored shared savings and 11% said employers or health systems.
That split is where the oversight question turns into a compensation question. If review is the service patients value, and employers measure AI’s return in added volume, the minutes spent checking output come out of the same day that now carries a higher visit target.
Review also carries a reputational cost that no productivity metric captures. A Penn State experiment with 135 participants, described in Medical Economics, found that when an AI second opinion disagreed with the doctor, patients rated the doctor as “lazier,” even though the doctor’s advice had not changed. The author argues that how disagreement is presented is a governance decision, and that the risk currently lands “on the clinician by default.” A physician who overrides the software can lose standing with the patient for doing exactly the job the patient said they wanted done.
The youngest physicians are the least confident. Only 55% of those under 30 said AI had raised or held steady their job security, against 90% of physicians in their 60s, and the younger group will spend the longest careers supervising these tools.
Two-thirds of physicians in the compensation survey said those who learn AI will have a meaningful earnings advantage in the next 12 months, yet only 23% expect AI to raise their own pay within a year. The gap between those numbers is roughly the gap between a skill employers will want and a skill they have started paying for.
The number that will decide the next contract cycle
The figure to watch is Doximity’s 28%: the share of physicians who say their organization has clear AI policies and approvals. Patients have said, by a margin of 56% to 9%, that physician review is the condition of their trust. The AMA has said the license, the title and the liability for mandated tools belong to people and organizations, not software. Until employers write those expectations into policy, training and contracts, the review patients are counting on will keep happening on physicians’ own time, and the first systems to define and pay for it will have an easier recruiting pitch than the ones that leave it implied.
Sources
Patient Comfort With AI Diagnosis Falls From 56% to 9% When No Doctor Can Review the Decision – Medical Daily
Who can be licensed as a physician? Not AI – American Medical Association
AMA: Artificial intelligence can augment physicians but don’t call it doctor – American Medical Association
AMA Seeks Legal Protection for Doctors Who Reject AI Advice and Wants Vendors Liable for Faulty Tools – Medical Daily
More Than Half of U.S. Physicians Now Use AI In Their Practice Daily, But Few Have Clear Rules For It – Doximity via Business Wire
How Physicians Feel About AI and If Their Pay and Job Security Will Be Affected – The White Coat Investor
When AI disagrees with the doctor, patients trust the doctor less – Medical Economics