The FDA’s public comment period on its discussion paper exploring how to regulate generative AI-enabled medical devices closed October 19, with the American Hospital Association among the groups submitting detailed feedback urging a risk-based approach that protects patients while still supporting innovation, as the agency weighs how to adapt its existing device framework to tools built on large language models rather than traditional, narrower machine-learning algorithms.
Why Generative AI Breaks the FDA’s Existing Playbook
The FDA’s traditional medical device framework was built around tools with fixed, predictable behavior that could be validated once and then monitored for drift, an approach that works reasonably well for earlier-generation diagnostic AI models trained for a single, narrow task like flagging a specific type of tumor on a scan. Generative AI systems, by contrast, can produce open-ended, context-dependent outputs and are often updated continuously, making it considerably harder to apply a one-time clearance process that assumes the underlying model’s behavior stays essentially static after approval.
What the American Hospital Association Specifically Asked For
In its October 19 submission, the AHA called for a regulatory approach that is explicitly risk-based, meaning tools used for lower-stakes administrative tasks would face lighter scrutiny than those influencing direct clinical decisions, while also asking for clear expectations that developers can design against and alignment with the operational realities hospitals face when actually deploying these tools at scale. The AHA’s involvement reflects how directly hospital systems, not just AI developers, have a stake in how this framework ultimately takes shape, since unclear or overly burdensome rules could slow adoption of tools hospitals already believe offer clinical value.
The Discussion Paper’s Limited but Notable Status
Importantly, the document that just closed for comment is explicitly a discussion paper rather than draft or final guidance, meaning it represents an early, non-binding stage in what will likely be a multi-year process before any formal generative AI medical device framework takes effect. This slower regulatory timeline sits in tension with the pace at which hospitals and AI vendors are already deploying generative AI tools for tasks like clinical documentation, patient communication, and preliminary diagnostic support, often under existing, narrower device clearances or outside formal device regulation entirely.
How This Connects to the FDA’s Broader 2027 Priorities
The generative AI discussion paper is one piece of a broader set of FDA priorities reportedly planned for 2027, which include specific guidance on generative AI for mental health applications and on AI-enabled robotic surgical systems, alongside a public device list that reportedly already included more than 1,600 AI-enabled medical devices authorized for marketing as of September 2026. That volume of existing authorizations, achieved under the FDA’s current framework, underscores both how much clinical AI adoption has already occurred and how much is riding on getting the next-generation generative AI framework right.
Industry Voices Want Speed; Patient Advocates Want Caution
AI developers and some hospital administrators have pushed for a streamlined approach that avoids re-litigating safety questions already addressed for earlier AI device categories, arguing that excessive caution could leave the US lagging in deploying high-value generative AI tools compared to healthcare systems in other countries. Patient safety advocates, by contrast, emphasize that generative AI’s capacity for confidently stated but incorrect outputs, familiar from consumer chatbot contexts, poses distinctive risks in medical settings that justify a more deliberate regulatory pace even if it slows adoption.
What’s Next
The FDA will now need to synthesize feedback from the AHA and other commenters into either a formal draft guidance or a revised discussion framework, a process with no fixed public timeline but that will shape how quickly and under what conditions hospitals can deploy generative AI tools for genuinely clinical, rather than purely administrative, purposes. Expect continued advocacy from both hospital groups and AI developers as the agency works toward its stated 2027 guidance priorities.
Photo: NRCgov / BY via flickr
