On August 18, 2026, the U.S. Food and Drug Administration issued a discussion paper opening the first dedicated federal regulatory framework for generative AI-enabled medical devices. The FDA's Center for Devices and Radiological Health, through its Digital Health Center of Excellence, is seeking public feedback on risk assessment, premarket evaluation, postmarket monitoring, and the specific challenges of foundation models and agentic AI, with comments due under docket FDA-2026-N-7874 by October 19, 2026.
The paper is not a binding rule. It signals the direction the FDA expects a future GenAI-specific premarket and postmarket regime to take, and the agency frames it as a potential model for regulators around the world. For manufacturers of AI-enabled devices, the comment window is the lever to shape obligations before they harden into law.
What does the framework propose for GenAI device review?
The FDA outlines a possible two-axis framework for risk assessment that would inform regulatory expectations, then builds a premarket evaluation approach on the concept of competency assessment. Modeled at a high level on how physicians are trained and evaluated, competency assessment would combine non-clinical device benchmarking with clinical confirmation to test whether a GenAI-enabled device performs as intended before it reaches patients.
The paper then sets out several potential approaches to risk-proportionate postmarket monitoring, and dedicates sections to foundation models and agentic AI systems, the two capabilities that most distinguish GenAI devices from traditional software and earlier AI/ML devices. For each area the FDA poses targeted questions rather than fixed answers, meaning the final obligations will reflect what commenters submit now.
Who has to engage, and by when?
The deadline is October 19, 2026, via docket FDA-2026-N-7874 on Regulations.gov. The FDA expressly invites device manufacturers, clinicians, consumers, researchers and the public. The exposed commercial audience is large and well-defined: GenAI-enabled medical device manufacturers and software-as-a-medical-device builders, including Medtronic, Boston Scientific, J&J MedTech, Abbott, GE HealthCare, Siemens Healthineers and Philips, plus big-tech healthcare teams moving foundation models into clinical use.
Because the paper contemplates obligations on foundation models and agentic AI, the reach extends beyond traditional device manufacturers to any team deploying a general-purpose model inside a regulated device. Regulatory affairs and compliance leads should treat the comment period as the moment to contest scope, thresholds and evidence expectations, not a procedural formality.
How would competency assessment change premarket review?
Today, AI/ML-enabled devices reach the US market through existing 510(k), De Novo and PMA pathways, with modifications handled under the FDA's AI/ML-based software-as-a-medical-device framework. The discussion paper does not abolish those pathways; it proposes a GenAI-specific evaluation layer on top of them.
The competency-assessment model splits evidence into two stages: non-clinical device benchmarking, which tests the model against standardized tasks in a controlled setting, and clinical confirmation, which checks whether the device performs as intended with real clinical input. The two-axis risk assessment would determine how much of each stage a given device must satisfy. The table summarizes the proposed shift:
| Dimension | Current AI/ML device path | Proposed GenAI approach |
|---|---|---|
| Risk model | Existing risk classification under 510(k), De Novo or PMA | Two-axis risk assessment |
| Premarket evidence | Substantial equivalence or PMA evidence package | Competency assessment: non-clinical benchmarking plus clinical confirmation |
| Postmarket oversight | Existing device reporting and modification controls | Risk-proportionate monitoring, several options posed |
| Scope | Device-specific AI/ML model | Extends to foundation models and agentic AI |
What should teams do before the deadline?
First, map every GenAI capability in your pipeline, including any foundation model or agentic component, and assess it against the two-axis risk proposal. Second, identify where your non-clinical benchmarking and clinical confirmation evidence is thin, because that is what the FDA will ask for. Third, draft comments on the questions that would impose disproportionate burden, and submit them to docket FDA-2026-N-7874 by October 19, 2026.
Continuous, per-jurisdiction real-time monitoring surfaces a consultation like this the moment it publishes, before its obligations harden into binding rules. FDA leadership, including Acting Commissioner Kyle Diamantas, CDRH Director Michelle Tarver and DHCoE Director Rick Abramson, has tied the paper to the administration's priority to harness AI and accelerate innovative medical products to market, so the final framework is likely to favor speed over caution where the comment record supports it.
Take advantage of this real-time watch
Verify which of your devices fall under GenAI, calendar the October 19 deadline, and brief regulatory affairs, digital health R&D and legal now. This framework is still in formation, and the comment record will decide its final shape; Obsidian follows this docket from discussion paper to proposed and final rule.
