AMA Says AI Can’t Be Licensed as a Physician — and Vendors Should Carry the Liability

What happened. On 9 November 2026, the American Medical Association’s House of Delegates adopted a new policy on artificial intelligence liability at the AMA’s 2026 Interim Meeting in Orlando, Florida (announced 10 November 2026). The policy draws a bright line: AI cannot be granted physician licensure, professional titles or certification to practise medicine. AI tools belong inside physician-led teams, used at the direction of the treating physician — and when a clinical AI product produces an erroneous output, the physician who used it as intended, while exercising appropriate clinical judgement, should not be the one left holding the malpractice claim.

Who carries the liability. The AMA’s position is that liability for patient harm caused by a faulty clinical AI output should rest with the AI vendor, manufacturer or distributor — not as a malpractice claim against the physician who relied on the tool in good faith. The AMA will push for federal and state legislation to make that stick: vendor product-liability insurance for clinical AI, disclosure of failure rates and performance across patient populations, a ban on contract terms that shift liability onto physicians or health systems, and protections for physicians who decline to use an AI tool — or decline to follow its recommendation. Health systems and other deployers share responsibility too, for how they select, validate and monitor the AI they put in front of clinicians.

Why this exists now. The policy builds on the AMA’s clinical AI liability framework released in October 2026, and it lands as AI moves from pilot to everyday practice: the FDA has now authorised more than 1,450 AI-enabled medical devices, and an AMA survey found a third of physicians cited a clear liability framework as the primary driver of AI adoption. Until now, the unanswered question — “who pays when the algorithm is wrong?” — has been one of the biggest brakes on clinical AI use. The AMA’s answer is that the answer should be the company that built and sold the tool.

What it means for doctors’ malpractice exposure. For physicians, the policy is a shield in the making, not a shield already in force — see the caveats below. If the AMA’s advocacy succeeds, a doctor who uses a clinical AI tool as intended and exercises independent judgement would face product-liability exposure sitting with the vendor, rather than a malpractice claim of their own. Three things still matter at renewal and in practice: (1) document the human-review step — prompts, outputs and the clinician’s independent assessment kept on file; (2) check whether your professional liability / malpractice wording covers work done with AI tools, and whether any AI exclusion has been attached; (3) read vendor contracts for liability-shifting clauses before signing — the AMA wants those banned, but they exist today.

Caveats, do not drop. This is AMA advocacy and policy, not law — it creates no statute and changes no liability rule by itself. Physicians remain responsible for independent clinical judgement; using an AI tool does not remove that duty. The legislation the AMA wants has not been passed, and vendor product-liability insurance for clinical AI is a proposal, not a product on the market today. Figures in this post are from the AMA’s 10 November 2026 announcement and the trade coverage of it.

Costs. Professional liability typically runs about $25–$109/month for the professions we track (see the Small Business Insurance Cost Index 2026). No premium range exists for clinical-AI vendor liability cover yet — it is still a policy proposal.

Sources: AMA announcement; Becker’s; MedTech Dive.

Leave a Comment