On 29 October 2026 the American Medical Association (AMA) published a new issue brief on liability for clinical AI, setting out who should answer when an AI tool used in care causes patient harm. The framework comes from the AMA Center for Digital Health and AI, and it starts from a simple principle: responsibility should follow authority, access to information, and the ability to prevent or reduce harm. Existing liability rules, the AMA argues, were not designed for tools where physicians, developers and deploying organisations each hold a different share of control.
Three Groups, Distinct Responsibilities
The framework divides the work into three groups, plus a fourth interest that must not be lost: patients. Developers and manufacturers should demonstrate and maintain the safety and effectiveness of their products, and deal with risks within their control. That includes giving users information about intended use, performance and limitations, so physicians can spot and respond to errors, rather than relying on warnings and disclaimers alone. Healthcare organisations and other deployers should be accountable for how AI systems are selected, implemented, governed and monitored once they are in live use. Physicians should be responsible for the clinical judgement they can meaningfully exercise, based on the system's intended use, its known limits, and the information available at the point of care. Patients, meanwhile, must keep meaningful avenues to seek accountability and appropriate remedies when harm results from a defective system or another actionable failure.
The Problem With the Current Setup
The AMA's concern is that physicians are currently being asked to act as a safety net for failures they may not be able to catch. Jennifer Goldsack, head of the AMA Center on Digital Health and AI, told Healthcare Dive that physicians may be held accountable for risks they cannot control and, in some cases, may not even be able to identify. The brief gives a plain example: a physician builds a treatment plan on a clinical decision support tool's recommendation, the tool misses critical data, and the physician has no visibility into that error. In that case, the AMA says, the physician should not carry the blame for the harm that follows. The brief also flags ambiguity over when product liability applies to AI, contract terms and disclaimers that developers can use to distance responsibility from risk control, and unsettled questions about whether deploying organisations carry AI-specific duties such as local testing or outcomes monitoring.
A Training Gap Behind the Debate
The framework did not appear out of nowhere. An AMA survey released earlier in 2026 found that a third of physicians cited clear liability frameworks as the main driver of AI adoption. The same survey, reported by Healthcare Dive, found that a quarter of physicians had received no AI training at all, while 92% were interested in more. Under the framework, training is part of what a deploying organisation owes physicians before it asks them to use a tool, alongside evaluation before rollout, monitoring for models that become less accurate over time, and a process for responding to adverse events. The AMA also says physicians should be able to decline an unproven system without penalty from their organisation.
Why This Matters for Healthcare Organizations
For hospitals, health systems and practices, the message is practical rather than legal theory. Before a clinical AI tool reaches the exam room, the organisation should be able to say why it chose that tool for that use, what evidence supports it, and what its limits are. Staff need training that covers those limits, not just how to click through the interface. Once the tool is live, someone needs to watch its performance, catch drift, and act when an adverse event is reported. Contracts need the same care as the clinical evaluation: a disclaimer that shifts all risk to the physician does not, in the AMA's view, make the tool safe. None of this changes the law on its own, but it gives leaders a checklist to use in procurement talks and governance meetings, and it gives physicians a clearer basis to ask questions before they sign off on an AI recommendation.
Analysis: Shared Control Means Shared Accountability
The most important word in the framework is not liability, but alignment. The AMA is not arguing that physicians carry no responsibility, and it is not arguing that developers carry all of it. It is arguing that accountability should track control, and that shared control means shared accountability, without any party's duties disappearing just because another party also bears some. That is harder to run in practice than a simple rule, because it requires evidence about who knew what, who could have acted, and what information was available at the time. Watch for how this thinking shows up in procurement contracts and in FDA feedback on generative AI devices.
Clinics reviewing how AI tools fit into front-desk and patient communication work can explore MedClinicX's AI patient portal, see how after-hours calls are handled on our AI medical receptionist page, or explore the demos and request a quote for their own practice. For related coverage, see our earlier article on hospitals telling FDA that AI developers should own post-market AI safety monitoring.
Sources and References
- American Medical Association — Liability for Clinical AI, AMA Center for Digital Health and AI, 29 October 2026: https://www.ama-assn.org/practice-management/digital-health/liability-clinical-ai
- Fierce Healthcare — Who's responsible for clinical AI? AMA has some thoughts, 30 October 2026: https://www.fiercehealthcare.com/ai-and-machine-learning/whos-responsible-clinical-ai-ama-has-some-thoughts
- Healthcare Dive — AI vendors, health systems join physicians in liability for clinical AI, 2 November 2026: https://www.healthcaredive.com/news/ai-vendors-health-systems-join-physicians-in-liability-for-clinical-ai/873492/
Reporting based on the sources above, accessed 3 November 2026. The AMA page was read directly, and the Fierce Healthcare and Healthcare Dive reports were read directly and cross-checked against it. MedClinicX did not conduct original reporting or interviews for this article. This article is for information only and is not medical or legal advice. Corrections: contact us via https://medclinicx.com/contact