Mitchell Berger

Dear Dr. Abramson: I write to make the following comments and suggestions in response to the above paper and request for feedback. The discussion paper correctly identifies many of the challenges associated with GenAI, including emergent behavior, open‑ended inputs, variable outputs, and reliance on third‑party foundation models. My comments below focus on definitional clarity, risk assessment, governance, patient‑facing versus health provider‑facing outputs, and additional confirmatory evaluation approaches. Please note that the views expressed are mine alone and not those of an agency, organization, or other individual(s). Full comment attached below. Sincerely, Mitchell Berger

Re: Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion
Paper and Request for Feedback, https://www.fda.gov/medical-devices/digital-health-centerexcellence/considerations-regulation-generative-ai-enabled-medical-devices-discussion-paperand-request [Docket no. FDA-2026-N-7874]
To: Rick Abramson, Digital Health Center of Excellence (DHCoE), Food and Drug
Administration
From: Mitchell Berger, (comments made in personal capacity), mazruia@hotmail.com. 8.25.26
Dear Dr. Abramson: I write to make the following comments and suggestions in response to the
above paper and request for feedback. The discussion paper correctly identifies many of the
challenges associated with GenAI, including emergent behavior, open‑ended inputs, variable
outputs, and reliance on third‑party foundation models. My comments below focus on
definitional clarity, risk assessment, governance, patient‑facing versus health provider‑facing
outputs, and additional confirmatory evaluation approaches. Please note that the views expressed
are mine alone and not those of an agency, organization, or other individual(s).
General comments: Other comments: E.g., for Questions 9 and 21:
Learning from international counterparts: FDA also learn lessons in developing standards
from to international regulatory counterparts, including the European Medicines Agency, Japan,
China, Canada, South Korea, and others to promote harmonized approaches to GenAI regulation
where feasible. Consistency across jurisdictions will reduce developer burden, improve patient
safety, and support global interoperability.
Emphasize patient-engagement and patient-centered design: FDA should encourage
developers of GenAI‑enabled devices to incorporate patients directly into AI development,
governance, monitoring, and evaluation. Patient engagement is essential for ensuring that GenAI
systems reflect real‑world needs, support autonomy, and avoid assumptions about patient
capabilities and likely responses to information.
Governance: Given the evolving and uncertain nature of GenAI risks, FDA should emphasize
the importance of strong governance structures within developing organizations. Developers
should be encouraged to seek legal, ethical, and clinical consultation when appropriate,
particularly for high‑risk or patient‑facing functions. In addition, FDA should incorporate ethics
expertise within its own review processes when appropriate, including staff trained in bioethics,
digital ethics, and responsible AI. GenAI‑enabled devices raise questions about autonomy,
transparency, fairness, privacy, biases, data-access and ownership and the boundaries of AIgenerated clinical guidance.
Pages 3 and 4: Definitions and context: FDA should collaborate with others in consistently
defining such terms as artificial intelligence and digital health: Terms such as ‘artificial
intelligence’ (AI) and ‘GenAI’ and ‘digital therapeutics’ should be consistently defined by FDA
in collaboration with others across governmental agencies. Similarly, researchers, academic
institutions, nonprofits and trade associations may have their own definitions of digital health, 1
Page
AI, etc.1 As developments of generative AI enabled digital mental health devices and other
technologies move forward, it may be helpful for FDA and others to develop and use when
feasible consensus definitions of such key terms as AI, generative AI, autonomous, digital
health, digital mental health and so on.2 To cite one example, the National Institute of Standards
and Technology definitions, as for foundation models, sometimes differ from those of FDA.3
Definitions used by NIST, NIH, ONC, and other agencies may differ from FDA’s own
terminology. As GenAI‑enabled technologies expand, consensus definitions will support clarity,
interoperability, and regulatory predictability.
Page 4, From Farm‑to‑Fork to Developer‑to‑User: Modernizing FDA Oversight for GenAI
Risks: FDA correctly notes that it “does not regulate GenAI as such; it regulates medical
devices, including GenAI-enabled devices.” This is accurate under 21 U.S.C. § 321(h), which
limits FDA’s authority to products meeting the statutory definition of a device. However, GenAI
introduces upstream risks that may not be adequately addressed through a device-only
framework.4
A useful analogy is food safety: FDA does not regulate how lettuce is grown, but FDA and the
US Department of Agriculture and others regulate safety across the entire farm‑to‑fork supply
chain because upstream conditions affect downstream risk. Similarly, if GenAI systems and
components will be used or adapted for FDA‑regulated software and devices, FDA will need to
work directly and collaboratively with other federal agencies, academia, and the private sector to
foster governance, safety, and effectiveness across the GenAI lifecycle. FDA should adopt a
developer-to-user perspective for GenAI that parallels the farm-to-fork approach for foods. 5
Pages 4–6 and Discussion Question 1: Limitations of the Two‑Axis Model: The two-axis model
discussed by FDA is a good starting point for discussion but does not capture such upstream
risks as hallucinations, algorithmic bias and incorrect data. While these axes capture autonomy
and harm severity, they do not account for upstream risks such as hallucinations, algorithmic
bias, incorrect or contaminated training data, or opaque model provenance. The model is also
primarily utilitarian, emphasizing consequences but not reflecting governance priorities, ethical
guardrails, or healthcare duties such as privacy, confidentiality, and transparency. In addition,
GenAI risks are not static. They may change as models are updated or not updated, as models
drift, as data degrades, or as context and use‑cases shift. A static two‑axis model cannot capture 2
Fatehi F, Samadbeik M, Kazemi A. What is Digital Health? Review of Definitions. Stud Health Technol Inform. 2020 Nov 23;275:67-71. doi:
10.3233/SHTI200696. Wienert J, Jahnel T, Maaß L What are Digital Public Health Interventions? First Steps Toward a Definition and an
Intervention Classification Framework J Med Internet Res 2022;24(6):e31921; Olivia A. Stein, Audrey Prost, Exploring the societal implications
of digital mental health technologies: A critical review, SSM - Mental Health, 2024(6): 100373, https://doi.org/10.1016/j.ssmmh.2024.100373
https://www.fda.gov/science-research/artificial-intelligence-and-medical-products/fda-digital-health-and-artificial-intelligence-glossaryeducational-resource; See e.g., ONC/ASTP’s glossary at https://www.healthit.gov/topic/health-it-and-health-information-exchangebasics/glossary and NIST’s at https://csrc.nist.gov/glossary
https://csrc.nist.gov/glossary/term/foundation_model
National Academy of Medicine; The Learning Health System Series; Elliott A, Krishnan S, Sarich T, et al., editors. Generative Artificial
Intelligence in Health and Medicine: Opportunities and Responsibilities for Transformative Innovation. Washington (DC): National Academies
Press (US); 2025 May 16. 3, RISKS OF GENERATIVE ARTIFICIAL INTELLIGENCE IN HEALTH AND MEDICINE. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK615587/;https://oecd.ai/en/genai/issues/risks-and-unknowns
National Academy of Medicine; The Learning Health System Series; Elliott A, Krishnan S, Sarich T, et al., editors. Generative Artificial
Intelligence in Health and Medicine: Opportunities and Responsibilities for Transformative Innovation. 2025. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK615587/
these dynamic, lifecycle‑dependent risks, underscoring the need for a more expansive and
adaptive regulatory framework.
Instead of a two-axis model, I would suggest a three-layer model:
Layer 1, Functional: Consequences (same as current graph/model) and Activity (same as current
graph/model).
Layer 2, Model Risks: addresses hallucinations, biases, data contamination, poor or incorrect
training data, susceptibility to adversarial prompts; and vulnerability to hacking, model
manipulation, or other security threats.
Layer 3, Lifecycle Risks: model drift, degradation, need for updates, stability.
Action-taking and Action-directing functions: The paper states that “CDRH is also
considering whether a patient facing informational function may not become any less directive
because it includes a ‘talk to your doctor’ or an ‘I am not a medical professional’ statement in
addition to the ‘action-directing’ information.” The assumption that patients will reliably seek
clinician input is unrealistic. The more likely outcome is that patients will simply consult another
AI model or tool to cross‑reference the first output. FDA should treat action‑directing outputs as
higher‑risk regardless of disclaimers.
Patient-facing informational functions versus health care provider-facing informational
functions and Discussion Questions 2 and 3:
Patient‑Facing Informational Functions: Patient‑facing functions should provide clear,
accessible, bottom‑line information that supports autonomy and shared decision‑making.
Interfaces may be simplified, graphical, or user‑friendly, but patients should have full optional
access to underlying data, uncertainty indicators, and model limitations.
Health Provider‑Facing Informational Functions
Provider functions may be more complex, detailed, and data‑rich. Clinicians may need
intermediate outputs, confidence scores, model rationales, and contextual factors that would not
benefit many patients. These functions support clinical judgment, diagnostic reasoning, and safe
integration of GenAI outputs into care (page 8). Outputs can be more directive because they are
interpreted within the clinician’s professional judgment and scope of practice.
FDA can consider the following factors when assessing patient‑facing versus provider‑facing
informational functions and the potential risks associated with each.
• Wording and tone: For patients, wording should be plain-language and non-directive.
For providers, wording should be technical and precise.
• Specificity: For patients, the output should be general with options and ranges. For
providers, outputs should be clinically-based with detailed parameters.
• Personalization: For patents, output should be tailored but non-directive. For providers,
output should support clinical judgment for specific cases. 3
Page
 • Transparency and data access: For patients: A simplified interface, but with full
optional access to underlying data, uncertainty indicators, and model limitations to
support autonomy and shared decision-making. For providers: Full access to all data
• Context and framing: For patients, Emphasis on choices, risks, benefits, alternatives,
and shared decision‑making. For providers, the focus can be on clinical factors, next
steps/workflows and potential comorbidities.
Safeguards to Mitigate Risk Without Underestimating Patient Capability
FDA could encourage safeguards that protect patients while preserving autonomy and access:
• Uncertainty indicators: Clear statements such as “These results may be uncertain” or
presentation of broad ranges.
• Shared decision‑making prompts. For example: “Discuss these findings with your
healthcare provider.”
• Reminders about rechecking results: Prompts indicating when information should be
rechecked or updated.
• Clarification that information presented is not a diagnosis: Reinforcing that outputs are
informational and do not replace clinical evaluation.
• Model update notifications: Indicators when a model has been changed, retrained, or
updated, so patients and providers understand when performance may differ.
Question 11: Additional Clinical Confirmatory Approaches FDA Should Add:
FDA’s list is strong but incomplete. Additional approaches for GenAI include:
• Hallucination & Adversarial Stress Testing: Inputs designed to provoke hallucinations,
misleading outputs, or unsafe reasoning.
• Bias, Fairness, and Representativeness Evaluation: Testing across demographic groups,
comorbidities, and underrepresented clinical conditions.
• Drift and update impact analysis: Pre‑ and post‑update testing, stability testing, and drift
detection.
• Testing in different clinical contexts and environments: Emergency departments,
inpatient, outpatient, specialty settings, and varied workflows.
• Human factors testing: Provider and patient comprehension, over‑reliance, safeguard
functioning, and clarity of uncertainty indicators.
Sincerely, Mitchell Digitally signed by Mitchell Berger
DN: cn=Mitchell Berger, c=US,
email=mazruia@hotmail.com
Berger Date: 2026.08.25 21:23:23 -04'00'

Mitchell Berger
Note: Please note that I am submitting these suggestions in my personal/private capacity. The
views expressed are mine only and should not be imputed to other individuals nor to any public
or private entity.
Page