US hospital and doctor groups tell FDA that developers should own generative AI device safety

The main US hospital, physician and telehealth groups have told the Food and Drug Administration that developers, not clinicians, should carry responsibility for the safety and performance of medical devices built on generative AI ¹. The American Hospital Association (AHA), the American Medical Association (AMA), the American Telemedicine Association and the Workgroup for Electronic Data Interchange (WEDI) filed comments on the FDA's proposed approach to regulating these devices, MedTech Dive reported on 23 October 2026 ¹.

What the provider groups asked for

The groups argue that monitoring generative AI tools such as chatbots and systems that synthesise medical information would put too heavy a load on providers, with small and rural practices hit hardest ¹. They want developers to own ongoing monitoring and the FDA to give providers a route for reporting problems ¹. The AHA said vendors and developers "must be responsible for the ongoing performance, maintenance and integrity of the tools they market" ¹. ATA Action, the telemedicine association's lobbying arm, wrote that "Manufacturers should remain responsible for the safety and performance of the devices they sell" ¹. Hospitals also told the regulator that limited insight into how the models work would make flaws hard for them to spot, even where governance for AI devices already exists ¹.

The groups broadly accepted the FDA's plan to grade risk by what a device does, from purely informational output, such as a prediction of a future cardiovascular event, up to tools that act entirely on their own ¹. They pushed for changes in two places. The AHA wants the likelihood of error or harm added as a variable, because a single function, such as advising on emergency care, can fail in both directions ¹. The AMA questioned the weight placed on human supervision, arguing that clinicians can make mistakes when reviewing AI output or come to rely on it too heavily, so the humans should be assessed as part of the risk ¹. WEDI asked for an adverse event reporting process and "nutrition" labels setting out what a tool does and its risks ¹.

What it means for the market

The filings set out where US buyers want liability to sit before generative AI devices reach wider clinical use. If the FDA follows them, the cost of post-market monitoring would move towards vendors, along with the exposure that comes with it. MedTech Dive notes that providers fear more malpractice claims while developers face negligence risk of their own, and that neither side wants the burden ¹. For suppliers, the AMA's argument matters because a human-in-the-loop design has been a standard way to position a tool as lower risk ¹.

The label request revisits an idea that has already been reversed once: the Biden administration required developers to produce similar cards, and the Trump administration rescinded that rule last year ¹. How far the FDA takes any of these points is not yet known.

Caveats

The account rests on a single trade press report of the groups' submissions, not the full comment texts, and it covers US regulation only ¹. Rick Abramson, director of the FDA's Digital Health Center of Excellence, said the agency plans to post draft guidance on generative AI-enabled medical devices in the coming months ¹.