AMA: 80% of U.S. Physicians Now Use AI Professionally — Double the 2023 Rate

By Hector Herrera | May 29, 2026 | Health

Eight in ten U.S. physicians now use AI professionally — a doubling in just three years — and most say it's making them better doctors. But adoption is outrunning governance, and that gap is becoming a patient safety problem.

The American Medical Association released new survey data showing 80% of physicians use AI in their professional work, up from roughly 40% in 2023. More than three-quarters of those physicians report that AI improves their ability to care for patients — a meaningful endorsement from a profession historically cautious about technology-driven disruption.

What the AMA Found

The AMA's survey documents rapid mainstreaming. AI in medicine is no longer a pilot program or an outlier's experiment. It is standard practice in an increasing share of clinical environments.

Key findings from the report:

• 80% of U.S. physicians use AI in their professional work — double the 2023 rate
• More than 75% say AI improves their ability to care for patients
• 40% report feeling both excited and concerned about AI in medicine
• Top concerns: patient privacy and protecting the integrity of the physician-patient relationship

The 40% "excited-and-concerned" split is the finding that deserves the most attention. Physicians aren't uniformly enthusiastic or uniformly resistant — they're cautious adopters who see the value and the risk simultaneously.

What Physicians Are Using AI For

The deployment landscape is visible from what health systems have publicly announced. Clinical AI today spans several categories.

Ambient documentation is the most widespread use case. AI listens to patient encounters and generates clinical notes, reducing the after-hours charting that drives physician burnout. Nuance DAX, Abridge, and Suki are deployed at scale across hospital networks nationwide.

Diagnostic support covers tools that flag imaging anomalies, interpret pathology slides, and surface differential diagnoses for complex cases where a second signal matters.

Prior authorization automation handles insurance paperwork, with some systems processing approvals in minutes rather than the days that manual review requires — a significant reduction in administrative friction.

Patient communication includes AI-drafted after-visit summaries and responses to routine follow-up questions through patient portals, freeing physicians and staff from inbox management.

At the enterprise level, both Mount Sinai and Mayo Clinic have moved beyond pilots to agentic AI workflows — systems where AI doesn't just assist but completes multi-step tasks autonomously. Mayo's platform coordinates care protocols across specialists. Mount Sinai's system flags high-risk patients for proactive outreach before a crisis develops.

The Governance Gap

The AMA data reveals a structural problem: adoption is outpacing the frameworks designed to govern it.

A physician at a well-resourced academic medical center using a rigorously validated AI diagnostic tool is having a fundamentally different experience than a rural practitioner using a consumer-grade AI assistant for clinical questions. The 80% adoption figure masks enormous variation in how AI is being used, by whom, and with what safeguards in place.

The AMA has published principles for augmented intelligence in medicine calling for:

• Transparency about how AI tools generate recommendations
• Physician oversight of AI-assisted clinical decisions
• Clear liability frameworks when AI contributes to adverse outcomes
• Patient disclosure when AI plays a meaningful role in their care

Most health systems haven't fully operationalized all four. Liability remains the most unresolved question. When an AI system recommends a treatment and the outcome is bad, who is responsible — the physician, the hospital, or the AI vendor? The answer isn't legally settled in most states.

State laws are beginning to fill the federal void. Georgia's SB 544, effective January 2027, will require human physician review of all AI-generated prior authorization denials. Idaho and California have enacted data privacy requirements for clinical AI tools. But these are state-by-state patches on a national-scale problem — and health systems operating across multiple states face inconsistent obligations.

Why the Doubling Matters

A 40-to-80% jump in three years is not gradual adoption. It's a behavioral shift at scale.

For context: electronic health record adoption took roughly a decade to reach similar penetration, and it required federal incentive payments and mandates to get there. AI reached 80% physician adoption without mandates. Physicians adopted because the tools reduced friction in their daily work — not because a regulator required it.

That organic adoption is both the good news and the problem. Tools that spread on usefulness alone spread without standardized vetting, training, or governance. Each hospital network, each physician practice, each telehealth platform is making its own decisions about which AI tools to deploy and how.

The AMA's concern about the physician-patient relationship cuts deeper than privacy. If patients believe their doctor's recommendation came primarily from an algorithm — and that belief will sometimes be accurate — it changes the trust dynamics that medicine depends on. Patients confide things to physicians they don't put in forms or apps. If they believe those confessions are being fed to AI systems, some won't confide at all.

What to Watch

The AMA data will likely accelerate calls for a federal clinical AI standards framework — something analogous to the meaningful use framework that drove EHR adoption, but focused on safety and transparency rather than simply deployment.

Watch for CMS guidance on reimbursement models for AI-assisted care. If Medicare begins paying differently for visits that used AI documentation or diagnostic support, it will create a financial incentive structure that accelerates both adoption and the governance conversations that should accompany it.

In the near term, the more immediate pressure is state legislation. Georgia's prior authorization law is a template. If it reduces payer disputes without increasing clinical errors, expect it to spread to a dozen states within two years.

The 80% number is the new baseline. The question now is whether governance can catch up before the first major AI-linked patient safety failure forces it to at crisis speed.