RSNA and ACR Seek Postmarket Monitoring for Generative AI Medical Devices

On August 18, 2026, the Food and Drug Administration released a discussion paper on how it might regulate generative AI medical devices and asked for public feedback by October 19 under docket FDA-2026-N-7874. Two radiology organizations responded. The Radiological Society of North America (RSNA) filed comments on the deadline day, and the American College of Radiology (ACR) also submitted comments. Radiology Business reported both filings on October 23.

On the point that matters most to departments already running AI, the two societies landed close together. Both want performance tracked in clinical use, not just at the point of FDA review. Clinical use is where radiologists read the output and where workflow changes show up.

The paper drew heavy interest. Radiology Business reported that FDA received more than 300 responses, citing Rick Abramson, MD, a radiologist who directs FDA’s Digital Health Center of Excellence.

What FDA’s paper on generative AI medical devices covers

The Digital Health Center of Excellence, part of the Center for Devices and Radiological Health (CDRH), is leading the effort. According to FDA’s August 18 announcement, the paper sought feedback on risk assessment, premarket evaluation, postmarket monitoring and other topics. It outlines a possible two-axis framework for assessing risk.

The most distinctive idea is a possible premarket approach built on “competency assessment.” FDA said it’s loosely modeled on how physicians are trained and evaluated, pairing non-clinical benchmarking with clinical confirmation. The paper also discusses risk-proportionate postmarket monitoring, foundation models and agentic AI, with targeted questions in each area.

FDA was explicit about the paper’s status. The CDRH discussion paper page says the document is for discussion only. It isn’t draft or final guidance, and it doesn’t propose or implement policy. The page also says the paper doesn’t address whether the approaches it describes fall within FDA’s existing legal authority. The societies are responding to an exploratory paper, not a proposed rule, and nothing commits the agency to any of their requests.

RSNA’s case for monitoring after deployment

As Radiology Business reported, RSNA argued that models changing after deployment create risks a one-time premarket review can’t address. The society wants continuous monitoring that reaches beyond performance to human-AI interaction, effects on workflow and whether the tool is being used appropriately.

RSNA also said manufacturers, not radiologists, must remain accountable. Its practical requests follow from that position: timely notice when a model changes, ways for users to report performance problems, and enough transparency to support local customization. Each of those requests has a counterpart at the site level. Someone has to receive the change notice, recognize a tool that’s underperforming and file the report.

ACR wants risk tied to clinical function and autonomy

In its October 22 summary of its comments, ACR said it wants a risk-based approach built on clinical function, degree of autonomy and effect on patient care, with credit for qualified end-user oversight. It asked for flexibility for continuous improvement, paired with strong postmarket monitoring, and for FDA to work closely with radiology and other specialties.

Radiology Business’s account of the ACR letter adds more specific requests. ACR urged FDA to regulate an AI tool’s complete clinical function. It also asked that evaluation focus on clinically consequential failures rather than generic hallucination rates. And before FDA publishes draft guidance, ACR wants a radiology-focused public workshop or advisory committee meeting, or at least imaging-specific feedback.

Who receives notices and reports under the societies’ positions

The societies agree on monitoring, but they approach oversight from different directions. ACR wants credit in a risk-based scheme for qualified end-user oversight. RSNA, as reported, insists that accountability stays with the manufacturer.

Crediting qualified end-user oversight need not contradict RSNA’s request that manufacturers remain accountable. The practical question is what a site takes on. A site that receives model-change notices and files performance reports is doing some of the monitoring work, and ACR’s summary doesn’t spell out what its credit for qualified oversight would mean in practice. How FDA would divide those tasks between users and manufacturers, while manufacturers stay accountable as RSNA asks, is unresolved. Whether draft guidance, an advisory committee meeting or a radiology workshop will follow, and when, is also an open question.

About the sources: RSNA’s positions and the figure of more than 300 responses are as Radiology Business reported them. ACR’s positions come from its own October 22 summary and from Radiology Business’s account of the ACR letter; both rest on the same letter. FDA’s two web pages describe the discussion paper, its non-guidance status, the docket and the deadline.