Who Bears Liability When AI Makes Clinical Decisions?

The accountability gap

A cardiologist overrides an algorithm's diagnostic recommendation. The patient recovers. Had she followed the AI's guidance and the patient suffered harm, she—not the algorithm's developer, vendor, or health system—would face legal consequences.

This scenario captures a fundamental tension as artificial intelligence enters medicine: AI systems increasingly inform clinical decisions, yet physicians alone bear legal and professional responsibility for outcomes. Two cardiologists writing in STAT argue this asymmetry must shape how the medical profession approaches AI governance.

Why licensure requires a person

Physician licensure has never been solely about passing examinations. It establishes a framework for accountability rooted in legal precedent stretching back more than a century. The Supreme Court upheld medical licensure in 1889 specifically because clinical consequences require someone who can be examined, tested, and held accountable.

Subsequent case law reinforced that physicians carry nondelegable duties. In Canterbury v. Spence, courts held that the duty to disclose material risks belongs personally to the physician and cannot be satisfied by proxy. Pennsylvania's Supreme Court ruled in Shinal v. Toms that this duty cannot be handed to colleagues, nurses, or systems—it remains with the physician because only the physician can be held to it.

AI systems can now pass the United States Medical Licensing Examination at or near threshold levels. But passing a test confers a credential, not a profession. The profession is defined by decisions made under uncertainty and outcomes carried personally.

The fragmentation of clinical decisions

As AI enters clinical workflows, what was once unified in one person—the clinical decision—now fragments across multiple actors. The algorithm performs analysis. The vendor captures billing. The payer deploys tools that authorize care. The health system licenses products into workflows.

Each entity extracts value: billing codes, attribution, outcomes data, market position. Yet accountability remains whole and undiluted with the physician whose signature appears in the chart.

This pattern has precedent. When imaging moved to reading rooms, physicians signed reports. When electronic health records automated order entry, prescribing physicians retained duty. When telemedicine crossed state lines, physician-patient obligations followed. Technology evolves, but accountability stays with the clinician.

Why it matters

The debate over AI licensure often focuses on technical capabilities—whether systems can match human performance on standardized tests. But licensure exists because of an asymmetry patients cannot close: they cannot evaluate whether clinical decisions were correct. Professional accountability requires a person who can be deposed, sanctioned by peers, and held to fiduciary duties courts describe as "of the highest degree." As AI reshapes how clinical decisions are formed, the legal and ethical framework for who answers when outcomes go wrong remains unchanged. Physicians leading AI adoption must recognize they are accepting the same personal responsibility their predecessors bore—regardless of how many algorithmic intermediaries now inform their judgment.

Leading the transformation

The authors, both cardiologists working to expand cardiovascular care access, support AI's potential to detect, triage, and personalize care at scale. But they emphasize that physicians who will answer for outcomes must lead how AI enters patient care.

That requirement—bearing personal responsibility for what happens to patients—has always defined medical professionalism. It remains the price and the point of the profession.

These details were first reported by Afnan R. Tariq and Ami B. Bhatt writing in STAT.