American Association of Orthodontists (AAO) (Nathan Mick)

See attached file(s)

Acting FDA Commissioner Kyle Diamantas, J.D.
Food and Drug Administration
10903 New Hampshire Avenue
Silver Spring, MD 20993

Re: Considerations for the Regulation of Generative AI-Enabled
Medical Devices: Response to Request for Feedback, Docket number
FDA-2026-N-7874

Dear Commissioner Diamantas:

The American Association of Orthodontists (AAO) submits these
comments on the important topic of risk assessment, premarket
evaluation, postmarket monitoring, and use of generative artificial
intelligence (AI)-enabled medical devices in health care.

Earlier this year, AAO published a position paper outlining how AI
can be incorporated into orthodontic clinical use and practice. Our
responses below stem from evidence-based information in this paper,
as well as the clinical expertise and experience of our approximately
19,000 member orthodontists.

We offer these comments from the perspective of healthcare
providers. AI can hold tremendous potential to improve patient care,
streamline administrative processes, and expand the capabilities of
medical professionals. However, AI can also introduce risks to patient
health and safety, particularly when used to make clinical
recommendations or treatment decisions without appropriate
professional oversight.

The AAO supports a risk-based regulatory framework that encourages
responsible AI innovation while maintaining meaningful safeguards
for patient health and safety. We believe AI should support the work
of licensed healthcare professionals, not replace their clinical
judgment or responsibility for patient care. Regulatory requirements
should be proportionate to the risks presented by each technology,
particularly where AI influences clinical decisions or treatment that
may result in irreversible biological changes.

The AAO's experience addressing patient safety concerns associated
with direct-to-consumer orthodontic treatment models underscores the
importance of appropriate clinical oversight. The AAO has
successfully advocated for the enactment of laws in nine states
requiring an in-person examination prior to orthodontic treatment to
help ensure patients receive appropriate professional evaluation before
initiating care. Additional resources on these issues are available at
OrthoFacts.org.

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 The AAO is grateful for the U.S. Food and Drug Administration’s (FDA) solicitation of comments on this
important issue. The AAO also appreciates being part of the FDA’s official Network of Experts. We
welcome working with the FDA to help inform appropriate regulatory approaches to ensure our country
can maintain its role as a health care innovator, while also protecting the health and well-being of
patients.

We submit the following comments in response to several of the questions outlined in the Considerations
for the Regulation of Generative AI-Enabled Medical Devices: Request for Feedback.

Considerations for the Assessment of Risk for GenAI-Enabled Devices

1. Does the two-axis risk framework, organized around device activity and the consequence of
relying on an incorrect output, appropriately capture the dimensions most relevant to the risk of a
GenAI-enabled software function? If there are additional dimensions—such as the reversibility of
a resulting action, the availability of downstream safeguards, the time pressure of the deployment
setting, or the traceability of the output (i.e., to primary source materials)—that should be
represented in a risk framework, please describe and provide examples of how they should be
represented.

The AAO believes that the two-axis risk framework provides a useful foundation for assessing risks
associated with GenAI-enabled software. However, additional considerations should include the
reversibility of resulting clinical actions, the potential for irreversible biological changes, and the
availability of meaningful clinical oversight. In orthodontics, inappropriate tooth movement can result in
lasting harm to teeth and supporting structures. These considerations should inform the level of regulatory
oversight necessary to protect patient health and safety.

AAO has also recommended that delegation frameworks and oversight requirements are proportionate to
(1) the intended clinical function of GenAI-enabled software, (2) the risk to the patient, and (3) regulatory
classification of the software. This three-pronged approach complements the two-axis framework laid out
in the FDA’s discussion paper.

2. CDRH seeks input on how to account for the spectrum of GenAI-enabled informational functions
that vary in the degree to which they direct a user to a particular action (i.e., between “nondirective” and “action-directing”). What characteristics of an output—such as its wording,
specificity, personalization, or context—could be considered as modifiers of the risk associated
with an informational function, after accounting for the device’s overall functionality and
intended use? What additional information would provide manufacturers with sufficient clarity
and predictability around risk assessment for informational functions while recognizing that
directiveness may exist along a continuum rather than as a binary distinction?

No Response Included

3. CDRH seeks input on whether and when a GenAI-enabled function that results in delivery of
clinical information to patients, as opposed to HCPs, could present different or higher risks,
while also recognizing the potential benefits associated with improved patient empowerment,

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 engagement, and access to clinical information. What device characteristics, output features, or
safeguards might mitigate risks that could arise when a user lacks the domain knowledge to
independently evaluate an output, without unnecessarily underestimating patient capability?

The AAO supports the responsible use of GenAI-enabled tools to improve patient education,
understanding, and engagement. However, AI tools that provide patient-specific diagnoses, treatment
recommendations, or other clinical directives should operate under meaningful oversight by an
appropriately licensed clinician, consistent with the AAO's Human-in-Command (HIC) framework.
Without such oversight, patients may act on inaccurate or inappropriate AI-generated recommendations,
potentially delaying necessary care or causing preventable harm to teeth, periodontal tissues, and oral
function.

Additionally, the AAO has successfully advocated for laws passed in nine states requiring an in-person
examination before initiating orthodontic treatment to address patient health and safety concerns
associated with direct-to-consumer treatment models. Similarly, AI tools that deliver clinical
recommendations or treatment decisions without appropriate human-in-command (HIC) controls and
oversight by licensed healthcare professionals can put patients at risk, undermine established standards of
care, and diminish the critical role of professional licensure in protecting patients.

AAO recommends that clinicians disclose and document AI use when it meaningfully contributes to
diagnosis, treatment planning, or care delivery. Explanations should reflect the tool’s role, relevant
limitations, and the clinician’s ongoing oversight. When AI use materially affects treatment risks,
benefits, or alternatives, its role should be addressed as part of the informed-consent discussion.

The AAO supports transparency regarding the use of patient data in AI development, training, validation,
and retraining, consistent with applicable privacy requirements and principles of informed consent.
Developers and manufacturers should provide clinicians with clear information about relevant data
practices so that clinicians can appropriately inform patients and fulfill their professional responsibilities.

4. CDRH seeks input on whether and how the distinction between generalist and specialist
physicians might inform the assessment of risk for HCP-facing GenAI-enabled functions. Under
what circumstances, if any, might risk be affected when an HCP who lacks the relevant clinical
specialist knowledge receives information that falls within a specialist area of practice? What
device characteristics or safeguards might mitigate such risks?

The AAO believes that GenAI-enabled software functions should support the knowledge, expertise, and
clinical judgment of licensed orthodontic specialists. These technologies should not substitute for the
specialized training, certification, and licensure associated with orthodontic specialty practice.
This position is consistent with the principles reflected in the FDA’s Clinical Decision Support Software
guidance, which describes non-device CDS functions as supporting clinical decision-making without
replacing or directing the health care professional’s judgment. The guidance also emphasizes that the
intended user must be able to independently review the basis for the software’s recommendations and
exercise their own clinical judgment.

Applying these principles to orthodontic care, the AAO believes that meaningful clinical oversight
requires expertise appropriate to the care being provided. When evaluating outputs requires orthodontic

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 specialty knowledge, oversight by a licensed health care professional who lacks that knowledge is
insufficient. Accordingly, the AAO does not support GenAI-enabled software functions designed to
replace an orthodontic specialist’s expertise by providing patient-care directives to health care
professionals who lack the specialty knowledge necessary to independently evaluate those outputs.

AAO reaffirms its support for the HIC oversight model for GenAI-Enabled software in orthodontic care,
which entails tools being used only under the oversight of a licensed orthodontist of record. When tools
are used outside of this framework, including HCP-facing tools, AAO has identified many potential risks.
These include:

• Use beyond validated context of use;
• Inappropriate or unaccountable delegation;
• Absence of lifecycle monitoring and feedback loops;
• Breakdown of legal and ethical accountability between clinicians and developers;
• Breakdown of legal and ethical accountability between clinicians and organizations;
• Erosion of public trust;
• Undermining licensure-based practice; and
• Increased patient-safety risks.

Safeguards should include clearly defined intended users and required clinical expertise,
accessible explanations of the basis and limitations of outputs, and validation demonstrating that
intended users can appropriately evaluate and apply those outputs. When evaluation requires
orthodontic specialty knowledge, a defined pathway for review by a qualified orthodontic
specialist should be available before recommendations are used to guide patient care.

5. For multi-turn conversational GenAI-enabled devices that may migrate from providing “nondirective” information to “action-directing” information over the course of an exchange, how
should risk be assessed across realistic conversational trajectories? How could the intended use
of such a device be characterized when its behavior is emergent across a conversation?

No Response Included

6. For GenAI-enabled care escalation functions, CDRH is considering how the evaluation may
account for both under-escalation and over-escalation. How could manufacturers characterize
and weigh these two directions of error, given that they may not be commensurable and that
acceptable trade-offs may vary by clinical context?

AAO is committed to ensuring that patient care remains paramount in the GenAI-Enabled Technology.
One way to mitigate the concern of care over-escalation or under-escalation from these devices in the
orthodontic context is to ensure that these tools are overseen and handled by a licensed orthodontic
specialist under the HIC methodology.

A Competency-Based Approach for Premarket Evaluation of GenAI-Enabled Devices

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 7. Is the competency-based approach described above, i.e., device benchmarking followed by
clinical confirmation, a useful and appropriate framework for evaluating GenAI-enabled
devices?

The FDA’s competency-based approach under consideration aligns with AAO’s view that developers and
regulatory bodies should define a tool’s context of use at the point of development. AAO believes this
context of use should influence how the tools receive regulatory classification, continued oversight, and
risk assessment.

The need for authorized context of use is great, as AAO has written about the dangers that come from
tools that go beyond their original approved function. In addition to bypassing “required validation and
quality management process mandated under U.S. and international device regulations”, these tools run
the risk of increasing “irreversible treatment errors” that would greatly harm patients.1

AAO further believes that oversight should be proportionate to the amount of potential risk. Strict and
unnecessary oversight can limit innovation and have unintended consequences. The discussion paper
outlines how the competency-based approach would rely on the two-axis framework to decide on the
rigor of evaluation and premarket testing, as lesser risk may lead to a reduction in premarket testing.
AAO’s views align with this principle, as our research has outlined that low risk AI tools that have little
impact on clinical judgement need not be subjected to a full regulatory requirement.

8. How might the two-axis risk framework described in Section IV be considered within the
competency-based approach described above to help determine the level of evidence needed for a
premarket submission?

AAO believes that patient safety must remain at the center of conversations surrounding rigor in
premarket evaluation and inform the need for tailored regulations. However, the FDA must also mitigate
the risk of overregulation hampering innovation. The two-axis framework within the competency-based
model would allow lower risk AI tools, specifically ones that do not impact clinical judgement, to be
subject to lesser regulations within the premarket evaluation. AAO favors this lighter touch approach
within this context.

9. Would a benchmarking structure such as the one described above be likely to provide adequate
evidence of clinical knowledge, analytic capabilities, safety behavior, communication, and
generalizability to support a reasonable assurance of safety and effectiveness? Are there elements
that are missing, redundant, or inappropriately categorized? Are there externally developed
standards that could be leveraged?

No Response Included

10. CDRH recognizes that publicly available benchmarking assets may be subject to data
contamination, saturation, and limited real-world representativeness. How should a sponsor
establish that performance on a given benchmark predicts safe and effective real-world behavior
for the device’s intended use? What evidence should support the construct validity of a
Hopkins H, Bianchi J, Cevidanes L, et al. Responsible integration of artificial intelligence in clinical care: An American Association of
Orthodontists position paper. American Journal of Orthodontics and Dentofacial Orthopedics. 2026;170:299–321.e4, page 306.
https://doi.org/10.1016/j.ajodo.2026.04.008.

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 benchmark used to gate device evaluation, and what role should sponsor-developed benchmarks
play given potential concerns around independence and optimization to the test?

The AAO shares CDRH’s concerns regarding bias, data contamination, and the limited clinical
representativeness of benchmarks used to evaluate GenAI-enabled medical devices. Sponsors should
maintain clear documentation of dataset sources, patient characteristics, selection methods, bias
assessments, and known limitations. Orthodontists serving as the Human-in-Command should review this
documentation and supporting validation evidence to assess whether a tool is appropriate for its intended
clinical use and their patient population. This transparency is also essential for ongoing clinical oversight.

Benchmark performance alone should not be considered sufficient evidence of safety and effectiveness.
Sponsors should demonstrate that benchmark results correspond to clinically meaningful performance of
the final configured device on independent, representative clinical cases, with the rigor of evidence
proportionate to patient risk. Evidence supporting construct validity should show that benchmark tasks
measure the clinical knowledge, reasoning, and safety behaviors required for the intended use. This
should include review by qualified, independent orthodontic specialists, reliable scoring methods, and
clinically justified acceptance thresholds established before testing. Evaluations should assess clinically
consequential errors and performance across relevant patient subgroups.

Benchmark assessments must ensure that data is neither saturated nor contaminated, and has a real-world
representation in order to serve as an effective tool for prediction. Benchmarks should be reassessed when
performance approaches a ceiling that limits their ability to detect clinically meaningful differences or
safety failures. Continued monitoring should confirm that performance remains acceptable as the device
and its clinical environment change.

11. CDRH is considering that clinical confirmation for a GenAI-enabled device might not require a
prospective clinical study in every case, and has described above a range of approaches of
increasing rigor and patient exposure. How might a sponsor select and justify a confirmation
approach tailored to a device’s intended use and proportionate to the device’s risk profile? Are
there device types or risk profiles for which one or more of these approaches would be
insufficient or inappropriate? How might the anticipated distribution of real-world inputs be
taken into consideration? Are there other methods of clinical confirmation that might help inform
the evaluation of GenAI-enabled devices?

No Response Included

12. How should sponsors achieve statistically meaningful performance measurement for GenAIenabled devices? Where synthetically generated inputs supplement real patient data, how should
sponsors account for differences between the synthetic and real-world distributions when
estimating performance, and under what conditions, if any, is it appropriate to combine
benchmarking evidence and clinical confirmation evidence to support a single performance
estimate?

No Response Included

13. For which clinical domains, device functions, or subpopulations is synthetic data particularly
well-suited, or particularly inadequate, as a supplement to real world evidence? What safeguards

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 would mitigate the risk that synthetic data generated by models of the same class as the device
under evaluation reproduces the very performance gaps the evaluation is intended to detect,
particularly for underrepresented subgroups?

The AAO believes synthetic data should supplement real-world evidence only when its equivalence to
real patient data has been validated for the specific clinical task and population being evaluated. This
assessment should address clinically relevant features, relationships, and variability; visual realism or
similar aggregate statistics alone should not establish equivalence.

In orthodontics, synthetic data may be useful for testing robustness to controlled changes in image
quality, incomplete or conflicting inputs, and unusual but clinically plausible scenarios. These
applications can help identify potential failures before clinical deployment. Greater caution is warranted
when evaluating growth prediction, biological responses to treatment, long-term outcomes, or
recommendations involving irreversible interventions. Synthetic data are particularly limited when the
underlying clinical relationships are poorly characterized or the source data inadequately represent the
population of interest, including patients with rare craniofacial conditions or underrepresented
demographic and developmental characteristics. Generating additional synthetic examples should not be
accepted as evidence that these representation gaps have been resolved.

Sponsors should report subgroup-specific performance and uncertainty, compare findings from synthetic
and real patient data, and obtain additional clinical evidence where meaningful gaps remain. Claims of
safety and effectiveness should be limited to the populations and uses supported by that evidence.

While synthetic data can support responsible innovation, the AAO cautions against relying on synthetic
datasets that have not been adequately validated for their intended clinical use. Patient safety requires that
such data be appropriately documented, representative, and supported by sufficient clinical evidence.

14. For open-ended device outputs, how should performance comparators and acceptance criteria be
selected? When a panel of qualified clinicians serves as the comparator, how should the
applicable standard (for example, the standard of care versus the performance of a median
clinician in practice) be defined and justified? Should generalist or specialist physicians be used
as a performance standard? When should human-AI team performance, rather than the device
operating alone, serve as the basis for evaluation?

The AAO believes that GenAI-enabled devices used to inform orthodontic diagnosis or treatment
decisions should operate under meaningful oversight by a licensed orthodontist, consistent with the HIC
framework.

Human-AI team evaluations should compare clinicians’ performance with and without the device under
representative clinical conditions, including their ability to recognize, question, and override incorrect
recommendations. The supervising clinician must possess the knowledge and expertise necessary to
independently evaluate the device’s outputs. Standalone device testing should complement these
evaluations by identifying intrinsic errors and limitations. When safe and effective use depends on
clinician interpretation or action, human-AI team performance should be the primary basis for clinical
evaluation.

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 15. Are there ways in which performance might be assessed relative to the care, technology, or
course of action likely to occur in the absence of the device, rather than to the comparators
described in this section? What approaches might be used to identify and justify a comparator
such as unaided clinical judgment, delayed specialist review, or no intervention?

As AAO strongly supports the HIC methodology, we believe that GenAI-Enabled devices should be used
in conjunction with supervision from licensed clinicians. If tools were to be judged through comparators,
AAO would likely support that this judgment come from licensed clinicians and best practices.

16. Should independent third parties be involved in some or all aspects of a competency-based
approach, including device benchmarking and clinical confirmation? If so, in what ways might
qualified, independent third-party participation contribute to this approach, and what
qualifications and independence criteria should apply? Are there aspects of the assessment for
which third-party involvement would be impractical or inadvisable? What safeguards or program
design features would be critical to prevent third-party participation from limiting competition or
preventing innovation?

The AAO supports qualified, independent third-party participation in the competency-based assessment of
GenAI-enabled devices, with the extent of involvement proportionate to patient risk. Independent parties
can strengthen evaluation by maintaining secure test datasets, assessing data contamination and
representativeness, verifying benchmark methods and results, and conducting clinical adjudication and
human-AI team assessments. Independent confirmation is particularly important when outputs influence
diagnosis, treatment planning, or other consequential patient-care decisions.

Evaluation teams should collectively possess expertise in the relevant clinical specialty, AI evaluation,
statistics, and human factors. Assessments involving orthodontic specialty knowledge should include
qualified orthodontic specialists. Reviewers should be independent of the device sponsor and any
underlying model developer, with documented disclosure and management of financial, employment,
advisory, and competitive conflicts.

17. CDRH is exploring whether a competency-based approach could be applied to devices that
incorporate a variety of underlying model architectures, such as multimodal vision-language
models and generative or predictive world models. Are there aspects of the described approach
that might be ineffective or inapplicable to these devices? What additional or distinct
considerations might need to be addressed?

No Response Included

Postmarket Monitoring for GenAI-Enabled Devices

18. CDRH is considering whether it may be appropriate to accept greater premarket uncertainty
regarding a GenAI-enabled device’s benefit-risk profile through greater reliance on postmarket
monitoring. Under what conditions might such an approach be appropriate, and what
characteristics of a monitoring program would need to be in place to justify reduced premarket

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 evidence? Are there device types or risk profiles for which this approach would not be
appropriate?

The AAO supports risk-proportionate premarket evaluation and robust postmarket monitoring consistent
with the FDA's Total Product Lifecycle (TPLC) approach. Increased reliance on postmarket monitoring
may be appropriate for certain lower-risk applications. However, postmarket oversight should not
substitute for sufficient premarket evidence of safety and effectiveness when GenAI-enabled devices
influence consequential clinical decisions or treatments that may result in irreversible biological changes.
The AAO also supports continuous monitoring of model performance throughout the device lifecycle to
identify performance drift, emerging bias, and other risks that may affect patient health and safety. The
rigor of both premarket evaluation and postmarket monitoring should reflect the device's intended clinical
use and potential risks to patients.

19. Please comment on the potential approaches to postmarket performance evaluation, including
periodic re-benchmarking, sample-based clinician review, and performance degradation
monitoring. What additional approaches should CDRH consider, and how should the cadence
and triggering events for reassessment be determined?

AAO supports strong postmarket evaluation, as models have the capacity to change and worsen over time
(even if the models are subject to strict premarket evaluation). Specifically, AAO believes in the FDA’s
outline of performance degradation monitoring. Importantly, we believe that regular postmarket
monitoring of GenAI-Enabled devices is critical and must be done even without triggering events.
Monitoring frequency and reassessment requirements should be proportionate to the device's intended
clinical use, risk profile, performance history, and potential consequences for patient safety.

20. Please comment on whether the proposed approaches to postmarket monitoring can be facilitated
by machine-based supervisory agents. What considerations, including the evaluation and
reliability of the supervisory agent itself, should CDRH take into account for such an approach?

No Response Included

21. What roles might clinicians, healthcare institutions, professional societies, standards-setting
bodies, and other stakeholders appropriately play in postmarket monitoring, and how can these
roles be structured without diffusing manufacturer accountability?

AAO recommends that licensed clinicians, healthcare institutions, professional organizations, and
standards-setting bodies play meaningful, clearly defined roles in the postmarket evaluation of GenAIenabled devices. Manufacturers should remain accountable for device safety and performance, including
ongoing monitoring, investigation of reported concerns, validation of updates, and timely corrective
action. Collaboration with other stakeholders should support these responsibilities without transferring
them to clinicians or healthcare institutions.

Consistent with the human-in-command (HIC) approach, licensed clinicians should evaluate device
outputs within their clinical expertise, report suspected errors and unexpected outcomes, and review
relevant safety alerts and update information. Assessments requiring orthodontic specialty knowledge
should include qualified orthodontic specialists. Healthcare institutions should establish reporting and
escalation procedures and monitor performance within their clinical workflows. Professional

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 organizations, including the AAO, can contribute specialty expertise, clinician education, and
mechanisms for identifying recurring concerns across practices. Standards-setting bodies can establish
consistent terminology, performance measures, and reporting formats to support comparable and
actionable monitoring.

Manufacturers should maintain risk-appropriate mechanisms for drift detection, error tracking, and
performance alerts, with information accessible to clinicians and regulators. Monitoring plans should
identify responsible parties, escalation thresholds, and response timelines, including how clinician
feedback will be investigated and incorporated into corrective actions or validated updates. Manufacturers
should communicate clinically relevant model changes, retraining events, known limitations, and
performance outcomes, and provide public summaries where feasible while protecting patient privacy and
proprietary information. Regulators should incorporate clinician and professional-society feedback into
ongoing oversight. These arrangements should preserve both manufacturer accountability for the device
and the licensed clinician’s responsibility for patient-care decisions.

22. CDRH envisions that a manufacturer’s premarket competency-based assessment might serve as a
baseline against which post-deployment modifications could be re-evaluated. How might the
extent of re-benchmarking or other evidence be scaled to the nature and expected impact of a
given modification? For example, are there categories of change that might not significantly
affect safety or effectiveness of a GenAI-enabled device, or might be appropriately managed
within a sponsor’s quality management system versus requiring FDA premarket review and
authorization before implementation? Are there categories of changes that might be appropriate
for inclusion in a PCCP?

No Response Included

23. How might PCCP concepts or other change-control approaches be adapted for GenAI-enabled
devices when the nature or scope of future modifications cannot be fully prespecified?

No Response Included

24. For devices built on third-party foundation models, changes to the underlying model may be
initiated by the third-party model developer rather than the device manufacturer. How can a
manufacturer detect, evaluate, and respond to such changes in a timely manner? What
mechanisms—for example, contractual, technical, or through a PCCP—could provide reasonable
assurance that third-party developer-initiated changes do not compromise the safety or
effectiveness of the device?

The AAO believes that manufacturers should remain accountable for the safety and effectiveness of
GenAI-enabled devices regardless of who initiates changes to the underlying foundation model.
Manufacturers should establish contractual and technical safeguards that allow them to detect, evaluate,
and manage these changes throughout the device’s lifecycle. Consistent with the human-in-command
(HIC) approach, this transparency should support informed clinical oversight while preserving
manufacturer responsibility for detecting and validating underlying model changes.

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 Unexpected changes or performance failures should trigger timely investigation and predefined protective
actions, including clinician notification, restoration of a validated and supported version, or suspension of
the affected function with a safe alternative workflow.

Other Topics

25. Would voluntary Foundation Model MAFs be practical for foundation model developers to
provide and to keep current, and what content might they include? Given that participation would
be voluntary and that model developers may have limited incentive to disclose safety-relevant
information, what would make such a program sufficiently useful for premarket review? Are
there alternative mechanisms CDRH should consider for obtaining information about underlying
foundation models?

No Response Included

26. Are there additional considerations that inform the premarket and postmarket evaluation of
agentic GenAI-enabled devices, beyond those applicable to non agentic GenAI-enabled devices?
How should the elevated risk associated with autonomous multi-step action, tool use, and
reduced opportunity for human review be reflected in acceptance criteria and oversight?

AAO reiterates its strong support for the HIC methodology and the use of GenAI-enabled tools when
they are deployed under the supervision of a licensed orthodontist. The elevated risks associated with
autonomous multi-step actions, tool use, and reduced opportunities for human review are particularly
significant in orthodontics because tooth movement is a complex biological process that affects not only
dental alignment, but also bite function, chewing, speech, jaw movement, and long-term oral health.
Orthodontic treatment relies on the application of controlled forces that cause biological changes in the
tissues supporting teeth, requiring individualized clinical judgment before, during, and after treatment.
Without meaningful human oversight, AI-enabled systems may fail to identify clinical factors that can
only be assessed through an in-person examination, including periodontal health, jaw function, signs of
temporomandibular disorders, and other patient-specific factors that may affect treatment safety and
outcomes.

For these reasons, AAO believes premarket evaluation of higher-risk agentic systems should assess
complete action sequences, including errors that compound across steps and failures involving external
tools. Acceptance criteria should require demonstrated ability to stop unsafe actions, escalate concerns to
the supervising clinician, and obtain clinician approval before initiating or materially changing treatment.
Postmarket monitoring should track executed actions, clinician overrides, and safety-related failures,
while preserving meaningful professional review throughout treatment.

Human oversight is also necessary to monitor for unintended consequences that may arise during
treatment and require professional intervention. Improperly planned or monitored tooth movement can
contribute to adverse outcomes, including periodontal damage, root resorption, tooth loss in patients with
untreated periodontal disease, temporomandibular disorders, and changes to occlusion that impair
chewing function, such as posterior open bite. In addition, licensed orthodontists play a critical role in
educating patients, obtaining informed consent, assessing patient compliance, and evaluating clinical
findings that cannot be fully captured through automated systems alone. For these reasons, AAO believes
acceptance criteria and oversight frameworks for higher-risk AI systems should prioritize meaningful

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org
 clinician involvement and preserve opportunities for ongoing professional review throughout the course
of treatment.

Thank you again for the opportunity to provide feedback on the use of generative AI-enabled medical
devices in health care. As the FDA continues to evaluate appropriate frameworks for risk assessment,
premarket review, and postmarket oversight, AAO encourages policies that promote innovation while
maintaining strong safeguards for patient safety, clinical effectiveness, transparency, and professional
accountability. We appreciate the agency’s consideration of these comments and look forward to
continued engagement on these important issues.

If you have any questions or AAO can be of further assistance, please reach out anytime via
NMick@aaortho.org.

Sincerely,

Nathan Mick
Vice President of Advocacy, U.S. and Canada
American Association of Orthodontists

401 North Lindbergh Blvd. | St. Louis, Missouri 63141 | 314.993.1700 phone | 314.997.1745 fax | aaoinfo.org