Krishna Koka

See attached files

September 1st, 2026

Dockets Management Staff (HFA-305)​
Food and Drug Administration​
5630 Fishers Lane, Rm. 1061​
Rockville, MD 20852​
Submitted electronically via www.regulations.gov

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

Dear Dockets Management Staff:

Thank you for the opportunity to comment on the Center for Devices and Radiological Health (CDRH)
discussion paper. I am a medical student at New York Medical College with a background in biomedical
engineering and prior work in craniofacial device development, including three-dimensionally printed,
patient-specific devices for cleft care. I submit this comment in my individual capacity; the views
expressed are my own and do not represent my institution.

Summary
This comment concerns a category the discussion paper does not yet name: generative AI
(GenAI)-enabled functions whose output is not information but a physical, patient-specific, implantable
device manufactured at or near the point of care. Section III of the paper states that it addresses the
GenAI-enabled component of a device and not the entire device. For most products, scoping is
reasonable. For an implant it sets aside the interface where the risk concentrates: the handoff from
generated geometry into build preparation, additive manufacturing, postprocessing, sterilization, and
release. My recommendations, in brief, are that CDRH (1) add physical output and reversibility as an
explicit dimension of the risk framework; (2) treat the validated production system, not the design
function alone, as the unit of evaluation for these products; (3) couple change control across the
digital-physical boundary; (4) structure clinical approval and manufacturing release as separate, mutually
necessary gates; and (5) specify a proportionate manufacturing record that makes later investigation
possible.

Why this category warrants attention now
The category is foreseeable rather than hypothetical. On June 18, 2026, CDRH cleared K253116, a
patient-specific titanium cranial plate system produced by the 3D Medical Applications Center at Walter
Reed National Military Medical Center, reported as the first such clearance held by a point-of-care
institution. I make no claim that AI is used in that system; its public summary identifies none. Separately,
peer-reviewed deep-learning pipelines already generate cranial implant geometry from CT data. CDRH’s
2021 discussion paper on 3D printing at the point of care already describes a medical device production
system—scanner, design software, coded limits, materials, printers, and processing—that may itself be a
device. The present paper and that concept should be connected before a GenAI-enabled design function
is placed at the head of such a system.

Responses to specific discussion questions

Comment on Docket No. FDA-2026-N-7874 — Page 1
Question 1 (additional risk dimensions). An implant proposed by a GenAI function sits beyond the top
of the activity axis: the output is an object, not advice, and remediation after implantation means revision
surgery. Both reversibility and traceability, which Question 1 raises, should be represented. I suggest a
physical-output modifier that presumptively places any function generating implant geometry at the
high-consequence end regardless of how the output is worded. The paper’s reasoning for measurement
functions applies with equal force here: a surgeon can judge anatomic fit visually but cannot see internal
porosity, a stress concentration, or an inadequate fatigue margin, so the user cannot independently
evaluate the properties that matter most.

Questions 7 and 11 (competency-based evaluation and clinical confirmation). I support evaluating the
final configured device rather than the foundation model. For physical-output functions, the final
configured device should be understood as the validated production system: imaging inputs,
segmentation, the design function, build preparation, printer and material, postprocessing, sterilization,
and release. This is consistent with the design or performance envelope and worst-case testing described
in CDRH’s 2017 additive manufacturing guidance. Benchmarking should therefore include what might be
called a generative design envelope—prospectively defined limits on image quality, modifiable
dimensions, feature size, fixation and load assumptions, build orientation, and postprocessing—with any
generated geometry outside those limits flagged as requiring engineering validation rather than released as
a routine patient match. Clinical confirmation for these functions should include physical evidence
(dimensional verification, mechanical and worst-case testing, nondestructive evaluation where
appropriate), not only clinician adjudication of outputs. Retrospective evaluation on real patient imaging
is a natural first step.

Question 18 (reliance on postmarket monitoring). Accepting greater premarket uncertainty in
exchange for postmarket monitoring is not appropriate for functions that produce permanent implants.
Monitoring can detect a problem in this setting; it cannot remediate one without surgery. Where the harm
is irreversible, premarket assurance of the manufacturing route should not be traded against surveillance.

Questions 22–24 (scaling evidence to modifications, PCCPs, third-party model changes).
Re-benchmarking triggers for these systems should include changes to printer firmware, build parameters,
material specification or lot, and postprocessing—not only changes to the model—and should specifically
address combinations. A workable tiering: interface-only changes managed within the quality
management system; a model change that alters which geometries can be generated triggers
re-benchmarking against the design envelope; a model change coupled with a manufacturing-parameter
change triggers worst-case physical testing before release. Predetermined change control plan (PCCP)
concepts can accommodate this if the plan enumerates interaction triggers across the digital-physical
boundary. For third-party foundation models (Question 24), production systems should pin model
versions so that an upstream update cannot alter geometry entering the manufacturing route without
passing a release gate.

Question 21 (stakeholder roles without diffusing accountability). I recommend a two-signature
release. The surgeon confirms anatomy, indication, and clinical intent; a qualified release authority within
the manufacturer of record confirms conformance using design checks, process data, and inspection
results. Neither signature substitutes for the other. This keeps accountability with the
manufacturer—which, for point-of-care systems, is the institution holding the clearance—while making
the clinician’s responsibility explicit and bounded rather than treating visual review as proof of properties

Comment on Docket No. FDA-2026-N-7874 — Page 2
it cannot reveal. Institutions and registries can then support surveillance that links revisions, imaging
follow-up, and adverse events to the model, design, and manufacturing configuration in use.

Question 26 (agentic systems). Section VII.B notes that an action sequence resulting in control of
another device may itself meet the device definition. A design function that writes a build file to a printer
inside a production system is such a function: it takes an irreversible, high-consequence action. The
human-oversight checkpoints described for agentic competencies (Appendix A, A.1) map directly onto
the release gates recommended above and should be required, not optional, for physical outputs.

Traceability
Each implant should carry a proportionate digital manufacturing record: source imaging identifiers;
software and model versions and clinically material configuration; generated versus edited geometry;
final design-file identifier; printer, firmware, build parameters, and material lot; postprocessing and
sterilization records; release signatures; and outcome linkage. The device history record requirements
already incorporated through the Quality Management System Regulation provide the base; the
GenAI-specific additions are model and configuration identity and generated-versus-edited provenance.

Summary of recommendations
• Add physical output and reversibility as an explicit dimension of the risk framework, with a
presumptive high-consequence placement for functions that generate implant geometry.
• For physical-output functions, define the unit of premarket evaluation as the validated production
system, using a prospectively defined generative design envelope and physical confirmation
evidence.
• Do not extend the postmarket-reliance approach to functions whose outputs are permanent implants.
• Couple change control across model, firmware, build-parameter, material, and postprocessing
changes, with tiered triggers and pinned model versions.
• Require separate surgeon approval and manufacturing release, and a proportionate digital
manufacturing record linking configuration to outcome.
I appreciate CDRH’s attention to these considerations and would welcome the opportunity to discuss
them further.

Respectfully submitted,

Krishna Sai Koka, M.S., B.S.E.​
Medical Student, New York Medical College (submitted in an individual capacity)​
kkoka@student.nymc.edu

Comment on Docket No. FDA-2026-N-7874 — Page 3