AMA assigns liability for clinical AI to vendors and health systems as well as physicians

The American Medical Association has published an issue brief that splits responsibility for clinical AI three ways, Healthcare Dive reported on Nov. 2, 2026. Physicians remain responsible for clinical judgment, including interrogating AI outputs and dismissing recommendations when necessary. Developers are accountable for designing safe tools and for giving users information about intended use, performance and limitations, not just warnings and disclaimers. Deployers, meaning health systems, hospitals and physician practices, are responsible for selecting the right tool for the clinical scenario, evaluating it before asking physicians to use it, training clinicians, monitoring performance for drift and setting up processes for adverse events. The brief also says harmed patients should have pathways to recourse, which Jennifer Goldsack, head of the AMA Center on Digital Health and AI, said includes compensation.

The core argument is that physicians should not be the safety net for failures they cannot see. Goldsack offered an example: if a decision support tool omits critical data and the physician has no visibility into that error, the physician should not be held responsible for resulting harm. The AMA says disclaimers are not strong enough safeguards and proposes identifying where risk lies, who can prevent it and what information they need.

For organizations deploying these tools, the deployer column is where the exposure sits. The AMA's 2026 survey, as cited by Healthcare Dive, found a quarter of physicians received no AI training and 92% wanted more. The same survey found a third of physicians cited clear liability frameworks as the primary driver of AI adoption, which suggests clarity on accountability can speed uptake of tools meant to reduce administrative burden and burnout.

What the evidence does not show: this is a policy position from a physician advocacy group, not a court ruling, statute or regulation. It does not change existing malpractice law on its own. Separately, the AMA and other provider groups told the FDA that developers should monitor generative AI devices, citing strain on small and rural providers, which signals where provider advocacy is headed rather than where rules have landed.

The move: have legal, compliance and the CMIO map each clinical AI contract against the three-way split. Check whether vendors supply performance and limitation data rather than disclaimers, whether indemnity matches developer responsibilities, and whether your own pre-deployment evaluation, training records, drift monitoring and adverse event processes would hold up as a deployer. Closing those gaps protects patients and gives clinicians a clear basis for trusting the tools they are asked to use.

You are a compliance and legal advisor to the executive team of [your organization], a [type of organization, such as health system, hospital or physician practice].

Context: Healthcare Dive reported on Nov. 2, 2026 that the American Medical Association published an issue brief splitting responsibility for clinical AI three ways. Physicians remain responsible for clinical judgment, including interrogating AI outputs and dismissing recommendations when necessary. Developers are accountable for designing safe tools and giving users information about intended use, performance and limitations, not just warnings and disclaimers. Deployers, meaning health systems, hospitals and physician practices, are responsible for selecting the right tool for the clinical scenario, evaluating it before asking physicians to use it, training clinicians, monitoring performance for drift and setting up processes for adverse events. The brief says harmed patients should have pathways to recourse. The AMA's 2026 survey, as cited by Healthcare Dive, found a quarter of physicians received no AI training and 92% wanted more, and a third cited clear liability frameworks as the primary driver of AI adoption. This is a policy position from a physician advocacy group, not a court ruling, statute or regulation, and it does not change malpractice law on its own.

Task: Map each clinical AI contract against the AMA's three-way split and identify gaps in vendor indemnity and oversight duties.

Output: A scorecard for the executive team with one row per contract and these columns: vendor supplies performance and limitation data rather than disclaimers; indemnity matches developer responsibilities; our pre-deployment evaluation; our training records; our drift monitoring; our adverse event process; gap rating; recommended fix; owner. Follow it with a short summary of the top gaps.

My inputs: contracts in scope [list of clinical AI contracts and vendors], owners [names for legal, compliance and CMIO], [your current clinician AI training completion rate], [your next contract renewal dates].

Guardrails: Use only the facts above and the inputs I fill in. Cite Healthcare Dive and the AMA issue brief by name. Flag anything uncertain or needing legal verification. Use no patient data. Mark this as a draft for the team to check, and note that nothing here is legal or medical advice.