Stanford Mussallem Center for Biodesign (Sandra Waugh Ruggles)

See attached file(s)

318 Campus Drive, E100
Stanford, CA 94305
biodesign.stanford.edu

October 19, 2026

Kyle Diamantas, J.D.
Acting Commissioner
Food and Drug Administration
5630 Fishers Lane, Room 1061
Rockville, MD 20852

Subject: Docket No. FDA-2026-N-7874, Considerations for the Regulation of Generative AI-Enabled
Medical Devices: Discussion Paper and Request for Feedback

Dear Mr. Diamantas, Dr. Tarver, and the FDA Digital Health Center of Excellence Team;

We appreciate the opportunity to provide general comments on Considerations for the Regulation of
Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback issued by FDA
Aug 2026. 1

As individuals and scholars associated with Stanford Mussallem Center for Biodesign,2 we believe
that a disciplined approach to innovation across the continuum of care is the key to solving
healthcare challenges in the U.S. and across the globe. For more than two decades, we’ve
demonstrated that innovation can be taught, learned, practiced, and perfected. The Biodesign
innovation process has been adopted by more than 100 universities and training programs around
the world. Our trainees learn to observe patient needs and care delivery to identify the right
opportunities for transformative impact. Over 22 million people have been helped by technologies
invented during Stanford Biodesign training.3 From our position at the metaphorical and physical start
line for healthcare innovation, we strive for policy that supports patient access and impact while
balancing safety, effectiveness, and cost-efficiency.

As the Agency is aware, to date, there have been more than 1600 artificial intelligence enabled
medical devices (AIMD) authorized by the FDA,4 predominantly in radiology, and the pace continues
to climb swiftly each month. New products, and new product classes are created each month, and
one of the first to pass through this process with an AIMD product was one of our own trainees
through the company Viz.ai.5

With the publication of the Discussion paper and request for feedback, FDA is providing an important
forum to stakeholders as to how generative artificial intelligence (GenAI) enabled products should be
regulated. The potential benefit for patients, physicians, and the healthcare system as a whole is
enormous. These products hold the potential to make complex health conditions easier to
understand and manage, translate and summarize health information, and support patients and
https://www.fda.gov/media/194242/download
https://biodesign.stanford.edu/
https://biodesign.stanford.edu/our-impact.html
https://www.fda.gov/medical-devices/digital-health-center-excellence/artificial-intelligence-enabled-medical-devices
https://www.viz.ai/
 318 Campus Drive, E100
Stanford, CA 94305
biodesign.stanford.edu

patient decision-making. The wide range of interactions and output also hold the potential to bring
together both empirical data such as imaging, laboratory results and self-reported symptoms into an
integrated solution. However, the potential also presents challenges to FDA regulation. The
probabilistic output of these tools, coupled with the breadth of information used for training,
introduces new questions of intended use, risk, and benefit that were not previously fully anticipated
under the FD&C Act and existing FDA Guidance documents.

Over the last 2 years, the Digital Health Advisory Committee (DHAC) has convened to discuss the
challenges of regulating GenAI enabled devices.6 The summaries of these meetings demonstrate how
experts from developers, professional societies, academia and regulatory law are thoughtfully
considering the topic. In addition, international efforts are also underway to help create structure and
oversight throughout the product lifecycle for GenAI enabled devices. These efforts are helping
cultivate growth in this rapidly progressing field.

The challenge for FDA is to balance oversight authority with predictability, transparency and
speed, so GenAI enabled devices can fulfill their potential to improve patient care and access.

It is our opinion that FDA has within its current regulatory framework the tools necessary to
appropriately define the authorization pathways for GenAI enabled medical devices based on the
intended use and claims. Under the Policy for Device Software Function and Mobile Medical
Applications,7 FDA defines which products require FDA pre-market review. In addition, predetermined change control processes (PCCP) define initial and subsequent design changes for
software as a medical device (SaMD),8 and Quality System Management Requirements (QSMR)9 and
Medical Device Reporting (MDR)10 define the design, testing, and ongoing monitoring and reporting
requirements that all manufacturers must uphold. FDA can further encourage and support GenAI
enabled devices through the following actions:

1. Publish guidance defining the GenAI functions subject to FDA pre-market review.
Provide examples of how a risk-based approach to classification will be used to determine
when a proposed function carries high vs. low-to-moderate risk, or will be subject to
enforcement discretion. Iterate the guidance using public comment and provide examples in
guidance documents. FDA should maintain a function-based approach grounded in intended
use, clinical actions, risk/benefit analysis, and demonstrated performance.

2. Clearly define the boundary of FDA device jurisdiction for GenAI enabled functions
while preserving states’ authority over professional practice. FDA should not assume that
characterizations of AI systems as ‘practicing medicine’ provide an appropriate framework for
device regulation. Automating a task performed by a clinician does not, by itself, make the
technology a medical practitioner. Timely clarification of FDA’s device jurisdiction is
https://www.fda.gov/medical-devices/digital-health-center-excellence/fda-digital-health-advisory-committee
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/policy-device-software-functions-andmobile-medical-applications
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/predetermined-change-control-plansmedical-devices
https://www.fda.gov/medical-devices/postmarket-requirements-devices/quality-management-system-regulation-qmsr
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/medical-device-reporting-manufacturers
 318 Campus Drive, E100
Stanford, CA 94305
biodesign.stanford.edu

increasingly important as states develop their approaches to clinical AI under their authority
to license healthcare professionals of all types. A clear federal reference point can help
prevent inconsistent assumptions about device functions from becoming entrenched in
duplicative or conflicting requirements.

3. Use unique product codes to define specific GenAI functions that are to be regulated
similarly. We support FDA’s existing function-based approach to products containing both
device and non-device functions. FDA should preserve this distinction and clarify its
application to GenAI enabled products, including conversational components. The use of a
shared model or conversational interface should not, by itself, bring otherwise non-device
functions within the scope of a device submission. Non-device functions should be able to
change without new submissions or PCCPs where those changes do not alter the regulated
device function or adversely affect its safety or effectiveness. Further, GenAI categories
requiring general controls can be separated from those requiring more extensive special
controls.

4. Publish and iterate guidance documents for each unique GenAI product category.
Assist innovators to determine specifically what valid scientific evidence would be
appropriate for FDA to understand risk/benefit trade-offs. Because scientific evidence can
range from literature, bench testing and benchmark testing, to simulated use testing, realworld evidence, and even clinical data, FDA should identify the evidence that is sufficient for
market authorization with each intended use or indication.

5. Prioritize Least Burdensome principles when determining what testing is sufficient to
demonstrate a reasonable assurance of safety and effectiveness. The use of GenAI or
technological novelty alone should not determine the level of evidentiary burden. Premarket
evidence requirements should instead be proportionate to the intended function, reasonably
foreseeable failure modes, probability and severity of harm, risk/benefit analysis, and the
effectiveness of validated safeguards and human oversight. GenAI enabled devices should
not automatically require more burdensome clinical evidence where benchmarking, simulated
use, or clinical confirmation methods can provide reasonable assurance of safety and
effectiveness. The FDA Guidance on Least Burdensome Principles provides a framework for
decision-making.11

6. Avoid creating duplicative testing standards and steer away from mandatory use of
third-party evaluation. Selecting a limited number of approved third-party evaluators can
create a choke-point on innovation, and also create perverse incentives which can
inappropriately pick winners and losers. Instead, voluntary third-party evaluation, such as
ASCA-type accredited testing laboratories12 or the use of voluntary consensus standards,
could be advantageous for GenAI developers and FDA alike. Such approaches promote
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/least-burdensome-provisions-concept-andprinciples
https://www.fda.gov/medical-devices/division-standards-and-conformity-assessment/accreditation-scheme-conformityassessment-asca
 318 Campus Drive, E100
Stanford, CA 94305
biodesign.stanford.edu

consistency, predictability and efficiency in medical device review and can reduce regulatory
burden and foster quality.
7. Align post-market requirements with Least Burdensome principles. FDA should draw a
distinction between post-market activities supporting regulatory decision-making, complaint
reporting, and quality assurance monitoring and performance reporting that are customer
requirements. Hospital and health system purchasers of GenAI medical devices may require
additional monitoring and deployment features that go beyond what is needed for QMSR, and
a robust competitive market will attract additional innovation in this area. Avoid monitoring
frameworks that create perpetual or exploratory endpoints.
8. Define when human oversight provides an appropriate risk control. Clinician oversight of
GenAI functions is meant to ensure high-quality system output and low rate of errors.
However, human oversight should also be evaluated based on its actual ability to mitigate risk
rather than treated as a binary characteristic. Relevant considerations include whether the
healthcare provider has sufficient information, authority, and opportunity to independently
review the output and intervene before a potentially harmful action occurs. FDA has written
guidance to this end when defining the boundaries of Clinical Decision Support tools.13

FDA, as a regulatory body, has the opportunity to play a substantial role in defining how innovation
utilizing GenAI tools will make an impact on patient care and access. Without well-defined
expectations and pragmatic policy, GenAI health application developers will abandon their effort in
the US. FDA’s current risk-assessment structure, augmented by the consideration of the factors
outlined in FDA’s discussion paper specific to GenAI medical devices, can be successfully used. For
truly novel GenAI devices that require FDA review based on risk and without predicates, the use of
the existing De Novo pathway, with applicable Special Controls would not only accommodate new
devices, but would create clarity for future developers by providing a clear predicate that they can
build upon. Subsequent devices may then be authorized under the 510(k) pathway, demonstrating
substantial equivalence, and innovation can be accelerated thereafter. By creating guidance
documents that clearly define submission requirements, including likely pre- and post-market
evidence frameworks, FDA can ensure that developers see FDA authorization as an important step
towards commercialization rather than a regulatory barrier to be avoided.

Sandra Waugh Ruggles, PhD Nancy Isaac, MPH, JD
Directory of Policy Research Domain Director, Regulatory Policy
Stanford Mussallem Center for Biodesign Stanford Mussallem Center for Biodesign

Josh Makower, MD
Co-Founder and Director
Stanford Mussallem Center for Biodesign
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software