Key Takeaways
- • The FDA released a discussion paper on generative AI-enabled medical devices on September 1, 2026, requesting stakeholder feedback by October 19, 2026, but explicitly noted the paper is not agency guidance and represents only preliminary thinking on regulatory approach.
- • The FDA is notably departing from established terminology by omitting references to previously used distinctions between "locked" and "adaptive" algorithms, avoiding SiMD/SaMD nomenclature, and largely bypassing the concept of "clinical decision support"—signaling a potential shift in regulatory framing rather than fundamental principles.
- • The agency proposes a "competency-based approach" to GenAI device evaluation modeled on how human clinicians are credentialed, emphasizing structured assessments of underlying knowledge and reasoning rather than exhaustive scenario testing, requiring dramatic changes to existing regulatory paradigms and likely statutory amendments.
- • The FDA identifies a two-axis risk framework centered on consequence of incorrect output and the activity's level of independence, while distinguishing between patient-facing and healthcare professional-facing functions—concepts familiar from prior software device regulation but repackaged for GenAI context.
- • Implementation of the proposed regulatory model would fundamentally reshape device review processes by requiring complementary post-market evidence alongside pre-market data, moving away from purely pre-market evaluation for GenAI-enabled devices.
Summary
The FDA's recently released discussion paper on generative AI-enabled medical devices signals a potential recalibration of regulatory terminology and approach, though the agency stops short of committing to concrete policy changes. Published on September 1, 2026, the 26-question discussion paper invites stakeholder input by October 19, 2026, explicitly disclaiming that it represents binding guidance. The paper acknowledges both the "transformative promise" and "unique risks" of GenAI-enabled medical devices, but notably departs from decades of established FDA regulatory vocabulary.
Most significantly, the FDA appears to be phasing out or de-emphasizing longstanding conceptual frameworks. The paper conspicuously omits the agency's historical distinction between "locked" and "adaptive" algorithms—a core taxonomy in prior AI device discussions. It also sidesteps traditional terminology around "software in a medical device" (SiMD) and "software as a medical device" (SaMD), and minimizes references to "clinical decision support" beyond a footnote. This linguistic shift suggests the FDA may be attempting to break free from prior regulatory constraints, though the underlying risk principles remain substantially unchanged.
The proposed regulatory model centers on a risk-based approach grounded in a two-axis framework: the potential consequences of an incorrect output and the level of independence with which the software directs or takes action. The FDA also maintains its longstanding distinction between patient-facing and healthcare professional-facing functions, recognizing that each presents materially different regulatory considerations. Critically, the agency acknowledges that pre-market review alone may be insufficient for GenAI devices, signaling an openness to post-market evidence collection as a complement to traditional pre-approval testing.
The most transformative proposal involves adopting a "competency-based approach" inspired by clinical credentialing models. Rather than exhaustively testing every scenario a GenAI device might encounter, the FDA would evaluate devices through structured assessments of underlying knowledge and reasoning capabilities—mirroring how human clinicians are trained and evaluated. This departure from scenario-based testing would represent a fundamental change to device review paradigms and would almost certainly require statutory amendments to the FDA's current authority.
Stakeholder engagement through the October 19 deadline represents a critical opportunity for industry, patient advocates, and investors to shape the contours of GenAI device regulation. The breadth of the 26 discussion questions suggests the FDA is still in active deliberation on substantive implementation details, making this feedback window potentially consequential for firms developing GenAI-enabled medical technologies.
Most significantly, the FDA appears to be phasing out or de-emphasizing longstanding conceptual frameworks. The paper conspicuously omits the agency's historical distinction between "locked" and "adaptive" algorithms—a core taxonomy in prior AI device discussions. It also sidesteps traditional terminology around "software in a medical device" (SiMD) and "software as a medical device" (SaMD), and minimizes references to "clinical decision support" beyond a footnote. This linguistic shift suggests the FDA may be attempting to break free from prior regulatory constraints, though the underlying risk principles remain substantially unchanged.
The proposed regulatory model centers on a risk-based approach grounded in a two-axis framework: the potential consequences of an incorrect output and the level of independence with which the software directs or takes action. The FDA also maintains its longstanding distinction between patient-facing and healthcare professional-facing functions, recognizing that each presents materially different regulatory considerations. Critically, the agency acknowledges that pre-market review alone may be insufficient for GenAI devices, signaling an openness to post-market evidence collection as a complement to traditional pre-approval testing.
The most transformative proposal involves adopting a "competency-based approach" inspired by clinical credentialing models. Rather than exhaustively testing every scenario a GenAI device might encounter, the FDA would evaluate devices through structured assessments of underlying knowledge and reasoning capabilities—mirroring how human clinicians are trained and evaluated. This departure from scenario-based testing would represent a fundamental change to device review paradigms and would almost certainly require statutory amendments to the FDA's current authority.
Stakeholder engagement through the October 19 deadline represents a critical opportunity for industry, patient advocates, and investors to shape the contours of GenAI device regulation. The breadth of the 26 discussion questions suggests the FDA is still in active deliberation on substantive implementation details, making this feedback window potentially consequential for firms developing GenAI-enabled medical technologies.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.