The U.S. Food and Drug Administration is preparing draft guidance on generative artificial intelligence in medical devices, with a senior agency official saying the administration has made the work a top priority. Rick Abramson, director of the FDA’s Digital Health Centre of Excellence, said the agency expects to move toward draft guidance in the coming months. He made the comments on Oct. 20 during a panel at AdvaMed’s The MedTech Conference in Boston.
Abramson said the issue has the secretary’s and the White House’s attention. His comments came as the FDA reviewed more than 500 submissions to its generative AI discussion paper on possible regulatory approaches for these medical devices. The discussion paper is not guidance and does not establish new regulatory requirements. Instead, the FDA used it to gather early input on how to assess and monitor generative AI medical devices.
FDA moves toward formal guidance
The FDA’s discussion paper covers several questions that could shape future regulation. It examines possible approaches to risk assessment, evaluation before a device reaches the market and monitoring after deployment. The agency also asks stakeholders to consider other issues related to regulating generative AI medical devices. The feedback period closed Oct. 19. MedTech Dive reported that the FDA had received more than 500 submissions by Oct. 20. Abramson said the agency needs time to review that feedback before moving ahead with formal guidance. He did not give a specific release date. Instead, he said manufacturers should expect draft guidance within months. That timing puts the FDA at an important transition point. The agency has moved beyond simply asking whether generative AI belongs within medical device regulation and is now working toward a more defined policy approach.
Generative AI has a separate FDA track.
The FDA is developing its generative AI work alongside its broader medical device software program.
The agency’s FY 2027 guidance list includes draft guidance for generative AI conversational devices intended for mental disorders. A separate topic covering risk assessment and evidence generation for generative AI devices is listed as under construction. Abramson said the mental health topic follows earlier Digital Health Advisory Committee discussions involving mental health chatbots. He described the broader risk assessment and evidence topic as an umbrella guidance.
The distinction matters because the two projects are at different stages. The mental health topic is listed among the priority draft guidance topics, while the broader risk assessment document remains under construction. Neither document is a final set of requirements for manufacturers.
Existing AI guidance is also being updated.
The FDA is also working on broader guidance covering AI software functions. The agency’s fiscal 2027 agenda includes final guidance on marketing submissions and lifecycle management for AI-enabled device software functions. The agency issued the underlying document as draft guidance in January 2025 and includes recommendations for documentation and risk management throughout the device’s product life cycle. That work applies to a wider group of AI-enabled devices than the generative AI guidance now being developed. The difference is important. Generative AI systems can introduce issues that guidance designed for more conventional AI software does not address in the same way. The FDA’s discussion paper therefore considers specific questions around risk, evaluation and monitoring for generative AI devices.
The agency is also continuing work on predetermined change control plans, which can help manufacturers describe planned modifications to medical devices and how those changes will be evaluated.
Postmarket monitoring is a key issue.
One central question in the FDA’s generative AI discussion is what happens after a device reaches the market. The agency’s paper asks for feedback on postmarket monitoring because generative AI performance can depend on the data, context, and conditions in which it is used. The FDA wants to understand how manufacturers can identify and manage performance changes after deployment.
That concern also appears in the FDA’s wider AI work. The agency’s current framework for AI-enabled medical devices considers activities across the product life cycle, including development, validation, deployment, monitoring, maintenance and modification. The discussion does not mean the FDA has already established a new monitoring requirement for generative AI devices. This paper aims to gather outside input before the agency develops a formal policy.
The 2027 agenda covers more than generative AI.
Generative AI is only one part of the FDA’s broader medical device guidance agenda. The fiscal 2027 list also includes final guidance on AI-enabled device software lifecycle management, predetermined change control plans and robotically assisted surgical devices. CDRH is accepting comments on the overall guidance lists through Nov. 30.
TechAmerica.ai previously examined those broader priorities in its article on FDA guidance priorities, which covered the agency’s plans for AI software, generative AI and robotically assisted surgical systems. Abramson’s newer comments add another layer to that earlier picture. The FDA’s generative AI work is no longer limited to a discussion paper. After reviewing the feedback, the agency now expects to move toward a draft policy.
Manufacturers are waiting for clearer rules.
For medical device companies, the eventual guidance could affect how the FDA evaluates and supports generative AI products with evidence. The FDA has not yet published the final regulatory framework, so manufacturers cannot treat the discussion paper as a new set of binding requirements. The document is designed to gather information on possible approaches rather than announce the agency’s final position. That leaves companies monitoring several related FDA projects at once. The agency is working on broader AI software guidance while developing separate recommendations for generative AI devices.
Abramson’s comments suggest the generative AI work is moving to its next stage. But the agency has not committed to a specific publication date, and the eventual guidance could change after the FDA considers the feedback it has received. For now, the clearest development is that the FDA intends to move from discussion toward formal policy. More than 500 submissions are now part of that process, while CDRH prepares for draft guidance in the coming months.
