AI Passed the USMLE — But Doctors Keep the Liability
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- AI has performed at or near the passing threshold on the United States Medical Licensing Examination, but cannot bear personal responsibility for clinical decisions, according to the opinion's authors.
- Afnan R. Tariq, M.D., J.D. — co-chair of the SCAI Artificial Intelligence Task Force and a CPT and RUC adviser — and Ami B. Bhatt, M.D., chief innovation officer of the American College of Cardiology and chair of the FDA Digital Health Advisory Committee, argue that accountability remains "whole and undiluted" with the physician even as vendors, payers, and health systems each capture pieces of the clinical decision.
- The authors cite the Supreme Court's 1889 ruling in Dent v. West Virginia, upholding physician licensure on the grounds that clinical consequences require a person who can be examined, tested, and held accountable.
- A companion 1898 ruling, Hawker v. New York, went further: character is as important a qualification as knowledge, because a profession is defined by what its members owe to those they serve.
- Two later rulings — Canterbury v. Spence (D.C. Circuit) and Shinal v. Toms (Pennsylvania Supreme Court) — establish that the physician's duty to disclose material risks is personal and nondelegable, the authors note.
- The opinion traces a pattern: each prior technology shift in medicine — imaging moving to reading rooms, electronic health records automating order entry, telemedicine crossing state lines — left the clinician still signing the report and still bearing the prescribing duty.
Why it matters: For hospitals and vendors racing to deploy AI clinical tools, the legal ground is settled against shifting liability: four cited rulings from 1889 to the present hold the clinical duty personal and nondelegable, meaning physicians alone face depositions, board sanctions, and fiduciary claims when patients are harmed — regardless of which algorithm informed the decision.
