1. The AMA's AI liability push in brief
The American Medical Association (AMA) adopted new policy at its Interim Meeting in November 2026 calling for laws that protect physicians from liability when a defective clinical AI tool harms a patient despite proper use. Under the AMA AI liability approach, responsibility for errors caused by faulty software would fall on vendors and manufacturers, which would also have to carry insurance and disclose their products' limitations.
The move comes as artificial intelligence spreads rapidly through hospitals and clinics, from tools that read imaging scans to systems that draft clinical notes or flag patients at risk of deterioration. Legal rules for assigning blame when those tools get something wrong have not kept pace, leaving physicians uncertain about where their exposure begins and ends.
As the nation's largest physicians' organization, the AMA carries considerable weight in health policy debates. Its new position sets out what doctors want lawmakers to do, but it does not change the law by itself. Federal or state legislators would have to act before any of these protections take legal effect.
2. What problem is the AMA trying to solve?
The AMA is trying to answer a simple but unresolved question: who is responsible when a physician relies on clinical software that turns out to be wrong? Today the answer is murky, and the AMA argues that doctors risk carrying blame for failures they had no way to foresee or control.
In traditional medical malpractice law, a physician is generally judged against the standard of care, meaning what a reasonably competent doctor would have done in similar circumstances. That framework was built for decisions made by people. It fits less neatly when a recommendation comes from an algorithm whose inner workings the physician cannot inspect and whose training data the physician never saw.
Product liability law, by contrast, is the body of rules that holds manufacturers responsible when a defective product causes harm. Medical devices have long been subject to it. The AMA's position effectively asks that clinical AI be treated more like other medical products, with the maker answerable when the product itself is defective.
3. How the AMA AI liability framework assigns responsibility
The core principle is that liability should follow control. The AMA's Center for Digital Health and AI advanced a framework at the end of October that ties responsibility for clinical AI to each party's ability to understand and manage the system, and the association's House of Delegates then adopted formal policy at its Interim Meeting.
The approach divides responsibility among three groups: the developers who build AI tools, the healthcare organizations that buy and deploy them, and the physicians who use them with patients. Each is expected to carry the share of risk that matches its capabilities and its role.
Where a clinical error stems from a defective AI product and the physician exercised appropriate clinical judgment, the AMA says responsibility should rest with the vendor, manufacturer or distributor rather than with the doctor. The physician's protection, in other words, depends on having used the tool properly and having applied sound professional judgment.
The logic is intuitive. A software developer controls how a model is designed, trained, tested and updated. A hospital controls which tools it buys, how it configures them and how it trains staff. A physician controls how a recommendation is weighed against everything else known about a patient. Aligning liability with those spheres of control is meant to put the cost of a failure on the party best placed to have prevented it.
4. What the AMA wants from AI vendors
The policy is not only a request for protection. The AMA also wants AI companies to accept obligations in return for the central role their products are taking in care.
Under the AMA's position, vendors should carry adequate product liability insurance, so that money is available to compensate patients if a defective tool causes harm. They should also disclose the limitations of their products and share performance data, giving hospitals and physicians the information they need to judge when a tool can be trusted and when it should be treated with caution.
These requirements would change the commercial relationship between clinical AI developers and the healthcare system. Disclosure obligations would make it harder to market a tool without being clear about where it performs poorly, while insurance requirements would give vendors a direct financial stake in the reliability of what they sell.
For hospitals evaluating competing products, standardized disclosure of limitations and performance data could also make comparisons more meaningful. Without such information, purchasers often have to rely on vendor claims that are difficult to verify independently.
5. Keeping a path to recourse for patients
The AMA, led by President Willie Underwood III, has stressed that shielding physicians must not leave injured patients without a remedy. Patients harmed by clinical AI still need a way to seek compensation, and the association ties that recourse directly to public trust in the technology: if people do not trust these tools, they will not be used effectively.
The framework's answer is to redirect claims rather than eliminate them. Instead of pursuing the physician who relied on a defective tool in good faith, patients would look to the developers and deployers of that tool. That is why the vendor insurance and disclosure requirements sit alongside the physician protections in the same policy: the insurance is meant to ensure that someone with resources stands behind each product.
This balance is likely to be a focal point in any legislative debate. Patient advocates and insurers will want assurance that moving liability away from physicians does not make it harder for injured people to be compensated, while developers will scrutinize how far their own exposure would expand.
6. Where hospitals and health systems fit
The middle tier of the AMA's three-way split, the healthcare organizations that deploy AI, often receives less attention than developers or doctors, yet it plays a decisive role. Hospitals and health systems decide which tools enter clinical workflows, how they are configured, which patients they are applied to and how staff are trained to use them.
Those choices can determine whether a tool performs as intended. A model that works well in one setting may perform differently in another, for example when patient populations, equipment or documentation practices differ from those the developer used during testing. Under a framework where liability follows control, an organization that deploys a tool carelessly or ignores known limitations would be expected to answer for that decision.
This is another reason the AMA wants vendors to disclose limitations and share performance data. Hospitals cannot manage risks they are not told about. Better information at the point of purchase would allow deploying organizations to match tools to appropriate uses, set safeguards where performance is weaker and give physicians a realistic picture of what the software can and cannot do.
Taken together, the three tiers are meant to create a chain of accountability in which each party has both the information and the incentive to reduce harm within its own sphere.
7. AI as a tool, not a doctor
Running through the policy is the AMA's long-held view that AI should augment physician-led care rather than replace it. In the AMA's view, these systems are meant to support the clinician's judgment, not stand in for it.
Consistent with that view, the AMA opposes granting AI systems licensure, professional titles or certification to practice medicine. Licensing is the mechanism by which states decide who may practice, and it carries obligations of training, ethics and accountability. Keeping AI outside that system keeps the licensed professional at the center of clinical decisions, and of the legal responsibilities that come with them.
The stance also clarifies the liability picture. If AI tools cannot be licensed practitioners, they cannot bear professional responsibility on their own. Someone else must answer for their failures, and the AMA's argument is that, when the fault lies in the product, that someone should be the company that made it.
8. Why the issue is pressing for physicians now
Clinical AI tools are arriving in medical practice at a rapid pace. Many doctors, however, remain uneasy about relying on systems whose errors could end up in a malpractice claim with their name on it.
Clear liability rules are widely seen within the profession as a precondition for trust. If physicians fear that following an AI recommendation, or departing from one, could expose them to legal risk regardless of how carefully they acted, they may hesitate to use the tools at all, slowing adoption of technology that could improve care.
Under the AMA's approach, the legal test would focus on whether the physician's professional judgment was sound, rather than simply on whether the physician did or did not follow the algorithm. That is intended to give doctors room to use AI as an input while still applying their own expertise.
9. What happens next
Adopting policy is the first step; turning it into law is a much longer process. The AMA has said it will advocate for federal and state legislation that reflects its position, which means the next phase will play out in statehouses and in Congress.
Any such bills would likely face scrutiny from AI developers, insurers and patient groups, each with a stake in how liability is divided. Details such as what counts as a defective AI product, what proper use looks like and how much insurance vendors must carry would all need to be settled.
Until lawmakers act, the existing patchwork of malpractice and product liability rules continues to apply. For now, the AMA's policy serves as a clear statement of where the medical profession wants the line drawn as AI becomes a routine part of patient care.