USC Department of Regulatory and Quality Sciences (Eunjoo Pacifici)

We support FDA’s early engagement with stakeholders as it develops an approach to risk assessment, premarket evaluation, postmarket monitoring, and lifecycle oversight for generative artificial intelligence (GenAI)-enabled medical devices. From a faculty perspective, we focus on an enabling issue that cuts across these questions: how FDA and its stakeholders can build the human capital, regulatory science research capacity, and cross-sector infrastructure needed to implement and continually evolve the framework. We are pleased to attach our observations and recommendations.

Comments on FDA’s Discussion Paper

Considerations for the Regulation of Generative AI-Enabled Medical Devices

Docket No. FDA-2026-N-7874

Submitted by:

The Department of Regulatory and Quality Sciences

The University of Southern California

Mann School of Pharmacy and Pharmaceutical Sciences

1540 Alcazar Street, CHP 140

Los Angeles, CA 90079

Contact:

Eunjoo Pacifici, PharmD, PhD

epacific@usc.edu

Chair and Associate Professor, Department of Regulatory and Quality Sciences

We appreciate the opportunity to comment on the U.S. Food and Drug Administration’s
(FDA) discussion paper, “Considerations for the Regulation of Generative AI-Enabled
Medical Devices.” We support FDA’s early engagement with stakeholders as it develops an
approach to risk assessment, premarket evaluation, postmarket monitoring, and lifecycle
oversight for generative artificial intelligence (GenAI)-enabled medical devices. The
discussion paper appropriately recognizes that GenAI systems may differ from
conventional software and AI-enabled devices in important ways, including variable
outputs, reliance on upstream models, evolving capabilities, and challenges in monitoring
performance over time. FDA also makes clear that commenters may focus on the
questions most relevant to their expertise. From a faculty perspective, we focus on an
enabling issue that cuts across these questions: how FDA and its stakeholders can build
the human capital, regulatory science research capacity, and cross-sector infrastructure
needed to implement and continually evolve the framework.

The recommendations below respond directly to themes in the discussion paper to
address the enabling infrastructure behind those concepts. If FDA expects new forms of
competency assessment, there must be a workforce capable of designing and interpreting
those assessments. If independent third parties are to maintain benchmarks and conduct

USC DRQS Comments October 2026 1
evaluations, academic and other institutions must have the methodological expertise,
independence, and clinical environments to do so. If postmarket oversight is a shared
responsibility, durable mechanisms must exist for government, academia, industry,
healthcare systems, and patients to generate and interpret evidence together.

1. Build a GenAI Medical Device Regulatory Science Competency Framework

We recommend that FDA convene academia, industry, healthcare systems, and
professional societies to define a tiered competency framework for the professionals who
will develop, evaluate, regulate, deploy, and monitor GenAI-enabled medical devices. The
purpose would not be to establish another credentialing requirement, but to clarify the
knowledge and skills needed across different roles.

The competencies should be interdisciplinary. Depending on role, they may include
medical-device regulation and regulatory science; machine learning, GenAI, foundation
models, and agentic AI; clinical medicine and workflow; biostatistics and clinical research
methodology; human factors and human-AI interaction; cybersecurity and software
lifecycle management; ethics, bias, and health equity; real-world evidence; and
postmarket surveillance. Recent educational literature illustrates how broad this
competency landscape has become, with proposed AI competencies spanning technical
foundations, clinical applications, ethics, law and regulation, critical appraisal,
professionalism, and research and innovation.

A shared framework would give universities a clearer basis for developing curricula,
certificates, continuing education, graduate programs, and fellowships. It would also help
FDA and industry articulate the capabilities needed in future reviewers, regulatory
scientists, clinical investigators, and technical evaluators.

2. Create a Cross-Sector GenAI Regulatory Science Fellowship

We recommend a dedicated 12- to 24-month GenAI regulatory-science fellowship that
builds upon FDA’s existing Medical Device Fellowship Program. The defining feature should
be a structured cross-sector experience, not classroom training alone. Fellows could
spend time in an academic research environment, at FDA, in an industry setting, and,
where feasible, in an academic healthcare system.

The academic component could provide advanced training in GenAI, regulatory science,
clinical research, statistics, human factors, ethics, and real-world evidence. FDA
experience could expose fellows to regulatory review, benefit-risk assessment, regulatory
policy, and postmarket questions. Industry experience could provide practical exposure to

USC DRQS Comments October 2026 2
data governance, model validation, quality systems, software lifecycle management,
model updates, and change-control strategies. Healthcare-system experience would allow
fellows to examine clinical workflow, human-AI interaction, unexpected use patterns, and
real-world performance.

Each fellow should complete an applied regulatory-science project addressing an FDArelevant problem, such as benchmark validation, clinical confirmation, human-AI
interaction, postmarket performance, foundation-model dependency, or evaluation of
agentic systems. Such a program would build people who can work across technical,
clinical, and regulatory boundaries, rather than simply increasing general AI literacy.

3. Develop a Longitudinal Talent Pipeline

A fellowship should be one component of a broader talent pipeline that begins before the
fellowship stage and continues through professional development. A GenAI-focused
pathway could connect students and professionals across medicine, pharmacy,
engineering, computer science, public health, biomedical informatics, and regulatory
science.

The objective should be reciprocal literacy. Individuals trained primarily in AI should gain
sufficient understanding of clinical evidence and medical-device regulation to work
effectively in a regulated environment. Individuals trained primarily in regulatory science or
clinical disciplines should acquire enough technical AI knowledge to critically evaluate
models, evidence, and lifecycle changes. This interdisciplinary foundation can be
reinforced through short courses, research practicums, internships, competitions, and
faculty development as well as formal degree and fellowship programs.

Faculty development is essential. Universities cannot sustain a GenAI regulatory-science
pipeline without faculty who understand both the technology and the regulatory questions.
FDA-academic exchanges, visiting-scientist opportunities, industry-academic
appointments, and shared case-study development could help maintain current expertise
as the technology evolves.

4. Establish an Academia-Government-Industry Regulatory Science Consortium and
Testbed Network

FDA’s questions about benchmarks, clinical confirmation, third-party evaluation, and
postmarket monitoring point to scientific challenges that extend beyond any single
product. We recommend a precompetitive consortium bringing together FDA, academic
institutions, medical-device and technology companies, academic health systems,
professional societies, standards organizations, patients, and other relevant partners.

USC DRQS Comments October 2026 3
An academic research program could address many of the unresolved regulatory questions
surrounding GenAI-enabled medical devices that are difficult to evaluate through
traditional premarket testing alone. Academic investigators can study how clinicians and
patients interact with AI systems in real-world workflows, including whether users
recognize unsafe outputs, appropriately override recommendations, understand
uncertainty, and avoid both automation bias and unwarranted distrust of accurate outputs.
Academic programs could help validate benchmark methods, maintain protected test
datasets, coordinate independent clinician adjudication, assess subgroup performance
and human-AI interaction, and conduct multi-institutional clinical confirmation studies.
USC could also develop competency standards and training for third-party evaluators.

Academic regulatory science programs can support qualified independent third-party
evaluations by combining regulatory science expertise with access to clinical, engineering,
data-science, biostatistical, and human-factors capabilities. This role should complement,
rather than replace, manufacturer accountability, with transparent conflict-of-interest
controls, predefined protocols, secure data governance, and separation between device
development and independent evaluation.

The FDA has a strong history of advancing regulatory science through collaborations such
as with Centers of Excellence in Regulatory Science and Innovation (CERSI), Medical
Device Innovation Consortium (MDIC), FDA Collaborative Communities, and the Software
Precertification (Pre-Cert) Pilot Program, Building on these models, a dedicated academicFDA partnership could establish an interdisciplinary research program to evaluate and
refine regulatory approaches for GenAI-enabled medical devices. Research would analyze
FDA experience, identify policy gaps, and develop methodologies for risk assessment,
clinical performance, transparency, lifecycle management, and human-AI interaction.

The consortium could address a focused set of shared regulatory-science priorities:
developing, securing, updating, and validating benchmarks; determining when simulation,
retrospective assessment, silent deployment, prospective studies, or real-world evidence
are appropriate for clinical confirmation; evaluating human-AI interaction, uncertainty
communication, automation bias, and clinician override; assessing generalizability and
health equity; and detecting performance drift and other postmarket signals.

Academic research centers could serve as distributed regulatory science testbeds for
simulation, retrospective evaluation, silent prospective deployment, clinician adjudication,
human-in-the-loop studies, prospective investigations, and controlled real-world
monitoring. Academic health systems would serve as platforms to assess safety,

USC DRQS Comments October 2026 4
performance, and model drift Academic health systems can conduct comparative and
pragmatic clinical studies evaluating device-assisted care against current standards of
care across diverse clinical settings and populations.

This multi-institutional network could also test whether performance generalizes across
patient populations, workflows, and institutions. The resulting evidence would inform riskbased policies, standards, and oversight mechanisms that support innovation while
protecting public health. Universities and research networks are also well positioned to
establish standardized regulatory research environments to assess model updates, version
control, performance drift, and third-party dependencies under controlled conditions.
Academic health systems can conduct comparative and pragmatic clinical studies
evaluating device-assisted care against current standards of care across diverse clinical
settings and populations. Through multisite studies, they can generate evidence on
subgroup performance, health equity, language and digital literacy considerations, and
clinician experience levels. Independent academic researchers can strengthen
methodological rigor through validated endpoints, robust statistical methods, blinded
adjudication, and comprehensive assessment of clinically significant errors and safety
events. Importantly, academic research can help develop evidence-based frameworks for
risk classification, human oversight, postmarket monitoring, and lifecycle management of
GenAI technologies. The resulting regulatory science evidence can inform FDA guidance,
technical standards, and risk-proportionate evaluation approaches that support innovation
while ensuring patient safety, clinical effectiveness, and public trust.

Further research is needed on changes to upstream foundation models and agentic
systems that perform multi-step tasks or act through external tools. Key questions include
when model changes materially affect downstream regulated devices, how to study
cascading errors, where human intervention is needed, and how actions should be
recorded and audited. Collaborative academic research can address these precompetitive
methodological questions alongside sponsor-specific development.

5. Pair Workforce Development With Funded Regulatory-Science Research

We recommend that FDA connect talent development directly to funded regulatoryscience research. Cooperative agreements, fellowships, public-private partnerships, and
Collaborative Communities provide existing mechanisms that could be used or expanded
for this purpose.

Trainees should participate in research addressing the scientific gaps FDA has identified
rather than receiving AI education in isolation. Projects could address benchmark validity
and contamination, hallucination and clinically consequential error detection, uncertainty

USC DRQS Comments October 2026 5
calibration, subgroup performance, synthetic-data validation, human-AI interaction,
postmarket signal detection, model-update assessment, and agentic AI evaluation. The
result would be dual capacity: new regulatory-science knowledge and a trained workforce
able to apply it.

This approach also creates measurable outcomes. FDA and partners could track the
number and disciplinary diversity of trainees, cross-sector fellowship completion,
placement into regulatory-science roles, multi-institutional research studies, validated
methods and tools, peer-reviewed publications, and incorporation of findings into
standards or regulatory practice.

Conclusion

Academic institutions can support FDA’s emerging framework through independent
scientific evaluation, interdisciplinary training, and clinical research. Universities can
develop and validate broadly useful methods, academic medical centers can assess
technologies across diverse workflows and populations, and faculty can train future
experts while advancing regulatory-science research.

GenAI-enabled medical devices pose both regulatory-policy and capacity challenges. FDA
can set principles for risk assessment, premarket evaluation, clinical confirmation, and
postmarket monitoring, but implementation requires a workforce and research ecosystem
equipped to apply them as technologies evolve. We encourage FDA to pair regulatory
expectations with investment in competency development, cross-sector fellowships, a
longitudinal talent pipeline, a precompetitive academia-government-industry consortium,
and a distributed network of academic and healthcare-system testbeds.

USC DRQS Comments October 2026 6