The Unlearned Intermediary: Clinical Decision Support, the Fiction of Independent Review, and the Allocation of Liability for AI Error in Medicine
Year: 2026
Type: preprint
Abstract: Clinical AI regulation and clinical AI liability rest on the same physician. Federal law exempts much clinical decision support software from device regulation when the software is intended to let a clinician independently review the basis of its recommendations. License terms and the treatment of such software as a tool for doctors place liability for its errors on the reviewing physician, as the Supreme Court of Alabama did in Sampson v. HeartWise Health Systems Corp. In November 2026 the American Medical Association sought a liability shield for physicians and described cases in which a physician cannot verify the inputs behind a recommendation. A randomized study of 457 clinicians found that explanations did not protect diagnostic accuracy against a biased model, and a juror experiment found that rejecting AI advice gives physicians less protection than legal doctrine predicts. The learned intermediary doctrine returns the duty to warn to the manufacturer when providers cannot reduce the risk of harm. Exempt software falls outside the doctrine's definition of a device, and licenses reproduce its allocation without that condition. A shield for physicians alone would leave injured patients without a liable defendant. Conditioning both the regulatory exemption and contractual disclaimers on demonstrated reviewability, with point-of-care disclosure of inputs and liability for mandating institutions, would keep the patient's claim intact.
Authors: Jefferson, Josie, Velasco, Felix
Language: English
Subfield: Law
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7233569890
Cites:
Cited by:
Related to:
