Core Principle 1: AI Governance and the Human-in-Command (HIC)
In clinical practice, AI systems must function as supportive tools that enhance, but never replace, the expertise of a licensed professional. Regardless of type or regulatory classification, all AI technologies that support or influence diagnosis, treatment planning, or patient care must operate within an AI governance framework that preserves professional accountability and patient safety.
AI Governance
AI governance encompasses the system-level responsibilities of identifying, evaluating, and approving appropriate technologies; validating their intended use; integrating them into organizational workflows; and monitoring their performance across the practice or entity.4,13,16 Governance establishes the structures, policies, and accountability mechanisms that ensure AI systems are safe, effective, and aligned with patient care standards.
It establishes who is responsible for decisions about how AI systems are developed, approved, and used, and ensures that these technologies operate safely, transparently, and in alignment with professional, legal, and societal values. In private practice, governance is carried out directly by the owner orthodontist. In small group practices, governance functions may be shared among partners. In large group practices and dental service organizations (DSOs), governance functions such as procurement, contracting, and enterprise-level policy occur at the corporate level.7
Human-in-Command Oversight
The Human-in-Command (HIC) concept, adapted from global AI governance literature*, is defined here as the licensed professional of record.13,24,27 The HIC exercises comprehensive oversight at the point of care, extending beyond clinical decision-making or output validation. This role encompasses the authorization, supervision, and accountable use of AI technologies in clinical practice. The HIC determines when, how, and by whom approved AI systems are applied within clinical workflows and ensures that every use of AI occurs within the boundaries of licensure and the professional standard of care.18,28
HIC oversight begins with assessing whether the selected AI tools meet a quantifiable clinical need and verifying that they have been validated for their intended purpose. Delegation of AI-related tasks must account for the level of patient risk associated with each AI function, as defined by regulatory bodies such as the FDA and must respect the duties legally permitted for staff under state dental practice acts.6,13,22,27 Oversight must be implemented and monitored at the practice level by the HIC to ensure compliance with regulatory guidance, protect patient safety, and uphold the standard of care.13,16,24,27,29
*The concept of Human-in-Command (HIC) originates in technology governance and appears in international health policy literature. In this paper, it is applied specifically to orthodontics, where the HIC is defined as the licensed orthodontist of record, consistent with American Dental Association (ADA) policy and global expectations that accountability rests with qualified clinicians.
Relationship Between Governance and Oversight
The separation between organizational governance and clinician oversight is essential. Governance establishes what AI systems are approved for use and defines the conditions under which they may be deployed.7 Oversight ensures that, in every clinical encounter, the licensed orthodontist retains authority over how those systems are applied to patient care.24 By clearly distinguishing these roles, the framework preserves accountability, prevents inappropriate delegation of clinical authority, and maintains patient protection.27
Risks of Absent AI Governance and HIC Oversight
Without structured AI governance and HIC oversight, orthodontic practices face operational, legal, and ethical vulnerabilities. Without clear, enforceable oversight by a licensed orthodontist in the role of HIC, AI tools may be used in ways that violate scope-of-practice laws, compromise patient safety, and erode trust in the profession. The absence of well-defined AI governance allows clinical decision-making to drift outside the bounds of licensure and regulation, exposing patients, providers, and the profession to avoidable harm.7,13,16,24 The key risks include:
1. Use Beyond Validated Context of Use
Without a licensed clinician overseeing how AI tools are selected, integrated, and supervised, systems may be deployed outside their validated context. This increases the likelihood of diagnostic inaccuracies, unsupervised treatment planning, or prescription errors based on unverified or non-clinically validated outputs.30
2. Inappropriate or Unaccountable Delegation
Orthodontists are always responsible for the delegation of duties to staff. Without proper oversight, AI may enable tasks that extend beyond legally permitted duties under state dental practice acts. This can lead to unauthorized delegation of diagnostic or treatment-related decisions, bypassing the orthodontist’s required judgment.16,27
3. Absence of Lifecycle Monitoring and Feedback Loops
When governance structures are not in place, AI systems may be deployed without mechanisms for continuous monitoring, validation, and detection of model drift. This disconnect between intended use and real-world performance violates both FDA and IMDRF expectations for post-deployment surveillance and safe lifecycle management.20,22
4. Breakdown of Legal and Ethical Accountability Between Clinicians and Developers
When responsibility for tool selection, validation, and use is not clearly defined, accountability becomes fragmented, and liability may fall inconsistently between clinicians and developers. While the orthodontist, as the licensed professional, ultimately holds legal responsibility for patient care, the growing influence of AI systems on clinical outcomes means that inadequate governance can shift additional, often unrecognized, liability to the provider.7,31-33
5. Breakdown of Legal and Ethical Accountability Between Clinicians and Organizations
In larger practices or corporate governance models, AI systems are often selected, validated, or deployed at the enterprise level without the orthodontist’s direct involvement. Even so, the licensed clinician remains legally responsible for how those systems are used in patient care.
Without clearly defined governance structures and HIC oversight, orthodontists may unknowingly assume greater liability than intended, bearing responsibility for clinical outcomes influenced by decisions made at the organizational or corporate level. Clear alignment between enterprise governance, HIC oversight, and clinical application is essential to ensure patient safety and protect both clinicians and organizations from unmanaged risk.7,16,27
6. Erosion of Public Trust
Patients expect orthodontists to be the final decision-makers in their care. If AI tools are used without clear oversight, or if decisions appear automated or unaccountable, trust in both the technology and the clinician is diminished.
7. Undermining Licensure-Based Practice
Licensure defines who is legally authorized to diagnose, plan, and deliver orthodontic care. If AI tools direct patient care without orthodontist oversight as the HIC, it undermines licensure and risks normalizing “provider-less” decision-making, eroding one of the profession’s core safeguards for patient safety.
8. Increased Patient-Safety Risks
Patient safety is the ultimate casualty when AI governance and HIC oversight are absent or poorly defined. Without governance to validate systems, monitor performance, and set boundaries for use, and without HIC oversight to ensure appropriate application in patient care, the safeguards that anchor clinical safety collapse.
Errors in diagnosis, treatment planning, or appliance design may go undetected when clinicians rely on AI tools that have not been properly vetted or are applied outside their validated context. When governance and HIC roles are unclear, the feedback loop between system performance and clinical outcomes breaks down, removing the critical human judgment needed to protect patients from algorithmic errors, bias, and inappropriate recommendations.7,13,16
Call to Action
To protect patient safety, ensure ethical care, and preserve trust in AI-integrated orthodontic practice, licensed orthodontists must retain final authority over all clinical decisions involving AI technologies. This authority should be exercised through an AI governance framework in which the HIC role is filled by the licensed professional at the point of care. The HIC is accountable for the AI tools used in patient care and for how those tools are applied to support or direct clinical decisions within the bounds of licensure and the standard of care.
Without clearly defined governance structures and professional accountability, AI systems may be used outside their intended purpose or without adequate clinical oversight. Such gaps can lead to inappropriate delegation of clinical authority, erosion of accountability, and potential harm to patients.7,34,35
Leading global frameworks affirm that qualified professionals must remain accountable for the deployment and oversight of AI in healthcare. While terminology may vary, the shared principle is consistent: the use of AI in clinical care must remain under the authority and direction of a licensed professional.13,24,28,36
The AAO calls on dental boards, regulatory bodies, developers, and industry vendors to:
1. Establish Well-Defined AI Governance Frameworks
Every orthodontic practice or organization implementing AI in a clinical setting must maintain a clearly defined governance framework that outlines the selection of AI tools, their validation for clinical use, and the conditions under which they may be deployed. Governance frameworks must define the role each AI system will play in clinical workflows, ensure transparency in how decisions are made, and preserve clinician autonomy in determining when and how AI tools are used in patient care, particularly when their outputs influence patient-care decisions.
2. Require Human-in-Command (HIC) Oversight at the Point of Care
All AI tools used in orthodontic patient care must operate under the oversight of a licensed orthodontist serving as the HIC. The HIC is accountable for how AI tools are used to support or direct patient-care decisions and must ensure that their application aligns with licensure requirements, regulatory expectations, and the standard of care. Oversight must be active, documented, and proportionate to patient risk, ensuring that clinical authority, ethical responsibility, and liability remain with the licensed professional at every point of care.
3. Establish Clear Delegation Frameworks Based on Risk and Context of Use
The AI governance framework should clearly define how delegation is structured based on state dental practice acts and allowable duties for staff. Oversight requirements must be proportionate to the AI tool’s intended clinical function, risk to the patient, and regulatory classification. HIC oversight at the point of care determines appropriate delegation based on the tool’s authorized COU and the level of risk associated with it.
For lower-risk tools, tasks may be delegated to trained staff, provided they fall within the scope of duties permitted under applicable regulations and remain under the orthodontist’s supervision. Regardless of delegation, the licensed orthodontist retains responsibility for ensuring each tool is used in accordance with its authorized COU and the standard of care. This risk-based approach aligns with oversight principles outlined by the FDA, IMDRF, WHO (World Health Organization), and NIST (National Institute of Standards and Technology).6,13,16,24
4. Prohibit Direct-to-Consumer AI That Bypasses Licensure
State dental laws should explicitly prohibit the use of AI technologies to diagnose, prescribe, or manage orthodontic treatment outside the supervision of a licensed provider. Direct-to-consumer (DTC) AI models that deliver clinical recommendations or treatment decisions without orthodontist oversight violate scope-of-practice laws and undermine patient protection. These systems bypass established safeguards of licensure, professional accountability, and HIC oversight, presenting significant risks to patient safety and public trust in the profession.
5. Align AI Use With State-Level Delegation Rules AI systems should be regulated using the same principles that govern delegation to human auxiliaries. Just as clinical tasks may only be assigned to staff based on training, competency, and licensure, AI clinical tools must not be used to perform or influence decisions that exceed the legal authority of the licensed provider. Orthodontists must remain responsible for how AI-supported tasks are assigned, validated, and supervised within clinical workflows.27,2
Core Principle 2: Regulatory Alignment and Risk-Based Oversight
AI tools used in orthodontic care must align with the highest regulatory standards based on COU, clinical risk, and lifecycle oversight. Global regulatory bodies, including the FDA, IMDRF, WHO, and ISO, have endorsed a risk-tiered regulatory approach for AI/ML SaMD. These frameworks require oversight to be proportionate to the system’s intended use and potential for harm, rather than applying uniform regulation across all tools.13,20,22,24
Adherence to Global AI/ML SaMD Guidance and Standards
Building on the governance framework established in Core Principle 1, Core Principle 2 defines how regulatory alignment and risk-based oversight ensure accountability across the AI lifecycle. AI technologies intended for patient care must conform to global frameworks that govern how systems are designed, validated, classified, and monitored.
AI developers and industry vendors are responsible for meeting regulatory and quality-management standards, maintaining transparency, and marketing AI tools only within the boundaries of their approved COU.20,22,24 Following the AI governance framework laid out in Core Principle 1, the HIC is responsible for oversight at the point of care, verifying that any AI tool used in practice has been developed in accordance with applicable standards and is used within its authorized scope.
U.S. FDA
The FDA applies its TPLC framework to AI/ML SaMD, outlining expectations for premarket validation, postmarket surveillance, and the use of Predetermined Change Control Plans (PCCPs) to manage adaptive algorithm changes.5,19,22,37
International Medical Device Regulators Forum (IMDRF)
IMDRF guidance defines SaMD and establishes the risk-based oversight principles adopted by global regulators to classify and govern such software, including AI-enabled tools.8,24,38
ISO 13485
This international standard specifies quality management system requirements for the design, development, and production of medical devices, including software-based tools.20
Risks from Inappropriate Oversight for AI Tools
1. Overregulation That Discourages Innovation
Subjecting low-risk assistive AI tools, such as those used for administrative, educational, or visualization purposes, to the same regulatory requirements as higher-risk medical devices can inflate development costs and delay clinical access.22,24,39 Evidence suggests that disproportionate regulation may discourage innovation, particularly among startups and academic developers, without corresponding gains in patient safety.1,13 Proportional, risk-based oversight is therefore essential to safeguard both patient protection and technological progress.
2. Clinical Harm from Off-Label Use and Misclassification
Marketing or deploying AI tools outside their authorized COU can introduce significant clinical risk.15 When systems are promoted or utilized for functions they were not cleared to perform, particularly diagnostic or treatment planning tasks, they bypass required validation and quality management processes mandated under U.S. and international device regulations.5,22,39 This increases the likelihood of automation bias, misdiagnosis, and irreversible treatment errors.40 Off label deployment also deprives the clinician of cleared labeling and performance evidence needed to assess whether an AI output is accurate or appropriate, thereby undermining professional judgment and exposing patients to harm.41
Together, these risks underscore the need for a unified regulatory approach that protects patients while enabling responsible innovation.13,22,24,39 Oversight must remain clinically grounded, proportionally enforced, and anchored in professional accountability. The following Call to Action outlines the steps necessary to ensure that AI/ML based SaMD in orthodontics are governed by both human expertise and regulatory rigor.
Call to Action
To ensure that AI technologies used in orthodontic care are safe, clinically appropriate, and ethically deployed, regulators, developers, and clinicians must share responsibility for aligning tool development and use with the highest global standards.
The AAO calls on developers, industry vendors, and regulatory bodies to:
1. Define and Validate Context of Use at the Point of Development
All AI tools intended for use in patient care must include a clearly defined and validated COU. This definition must inform the tool’s regulatory classification, risk assessment, and lifecycle monitoring requirements.
2. Comply with Risk-Based Regulatory Frameworks Across the Full Product Lifecycle
Developers must adhere to regulatory standards such as the FDA’s TPLC framework, IMDRF SaMD guidance, and ISO 13485, which collectively define premarket validation, postmarket surveillance, change-control mechanisms, and, where applicable, Predetermined Change Control Plans (PCCPs) for adaptive systems.20,22,24
3. Market Tools Only Within Their Authorized Use
AI technologies must not be promoted or deployed for functions outside their approved regulatory scope. Off-label marketing increases risk to patients, undermines clinician trust, and may trigger legal or regulatory consequences.
4. Support Regulatory Harmonization and Transparency
To foster both innovation and public trust, regulators should work toward consistent international standards that define expectations for classification, explainability, performance monitoring, and ethical deployment.
5. Enable Proportionate Oversight to Promote Responsible Innovation
Low-risk AI tools that do not inform clinical judgment should not be subjected to full regulatory requirements. Regulators should continue to support scalable oversight models that reduce barriers to innovation while maintaining safety.
By ensuring that AI tools are accurately classified, transparently developed, and deployed only within their approved context, the orthodontic profession can accelerate access to responsible innovation while protecting patient safety and professional accountability.
Together, Core Principles 1 and 2 reinforce the dual obligation of clinicians and developers: to maintain human authority in clinical care and to align tool development with global regulatory expectations.
Core Principle 3: Trustworthiness and Transparency Across the Lifecycle
AI tools must be reliable, explainable, and consistent from development through deployment and postmarket use.6,22,28 For clinicians to use these tools safely, they must have visibility into how models are trained, validated, updated, and monitored over time.19,24,40,42
Trust is not earned by performance claims alone. It requires transparency across the full product lifecycle, clear documentation of training data sources, validation procedures, update history, and mechanisms for ongoing risk monitoring. Clinicians are not responsible for auditing algorithms or inspecting code, but they must be able to verify that any AI tool adopted into practice is supported by trustworthy development practices and vendor accountability.5,22,24,43
This expectation is reinforced in the American Dental Association’s (ADA’s) Standards Committee on Dental Informatics (SCDI) White Paper No. 110627, the American Medical Association’s (AMA’s) Principles for Augmented Intelligence in Health Care36, ISO/IEC (International Organization for Standardization/International Electrotechnical Commission)28, and guidance from the WHO29 and the U.S. Department of Health and Human Services44. These frameworks emphasize that trust in clinical AI must be built on evidence, transparency, explainability, and postmarket accountability, not assumptions or proprietary opacity.
To operationalize trust and transparency, AI tools must be developed, validated, and maintained in accordance with five foundational pillars. These pillars establish the developer’s obligations across the product lifecycle and define the verification responsibilities of the licensed orthodontist serving as the HIC at the point of care.
Five Pillars of Trustworthiness in AI Development
1. Dataset quality and bias mitigation
2. Rigorous clinical validation
3. Explainability and auditability
4. Continuous monitoring and risk management
5. Transparency and vendor accountability
A summary of these pillars and of trustworthy AI development, outlining the corresponding responsibilities of developers and the verification requirements for orthodontists acting as the HIC can be found in Table 1.
Table 1. Five pillars of trustworthy AI/ML SaMD development and corresponding roles of developers and orthodontists (HIC)a
| Pillar | Focus | Developer Responsibilities | Orthodontist (HIC) Verification |
| 1. Dataset Quality and Bias Mitigation | Use of reliable and representative data | Build models with diverse, well-documented datasets.Test for and reduce demographic or data bias.Keep clear records of how data were collected and prepared. | Review vendor reports showing dataset sources, diversity, and bias testing.Confirm the data reflect the tool’s intended clinical use. |
| 2. Rigorous Clinical Validation | Proof of safety and performance | Validate tools using real clinical data and risk-based testing.Report key metrics such as accuracy, sensitivity, and specificity.Recheck performance regularly after release. | Confirm the tool has been independently validated and cleared for its intended use.Review performance metrics and regulatory classification. |
| 3. Explainability and Auditability | Understanding and tracing AI outputs | Design tools that show how results are generated (e.g., heatmaps, confidence scores).Keep logs linking each output to its model version and dataset. | Ensure outputs are interpretable and confidence levels are visible.Retain the ability to review and override AI-generated results. |
| 4. Continuous Monitoring and Risk Management | Ensuring safety over time | Monitor model performance and update when data or standards change.Document updates, known issues, and retraining results.Alert users to any performance changes or risks. | Check that monitoring systems and update logs are available.Confirm version histories and alerts are communicated clearly by the vendor. |
| 5. Transparency and Vendor Accountability | Openness and responsibility | Provide clear labeling, documentation, and contact pathways.Disclose training methods, limitations, and known risks.Respond to clinician reports and maintain traceable records. | Require access to documentation before adopting the tool.Keep records of vendor communications, updates, and issue reports. |
a Sources: ISO22, NIST6, WHO28, IMDRF24, FDA5,29
Risks to Patient Care and Clinical Adoption if Trustworthiness Is Not Upheld
AI tools used in orthodontics, independent of their regulatory classification, must meet foundational standards for trustworthiness.6,22,28,29 When these safeguards are absent, tools may produce unreliable, biased, or opaque outputs that compromise clinical decision-making and expose patients and providers to harm.24,40,43 The following risks illustrate what can occur when AI systems are adopted without sufficient transparency, validation, and ongoing oversight5,8,13,22
1. Patient Harm from Inaccurate or Unvalidated AI Outputs
Poorly validated models, or those trained on low-quality or non-representative data, can produce inaccurate outputs, including false positives and false negatives, that directly lead to misdiagnosis or unsafe treatment decisions. In orthodontics, this may result in overlooked pathology, mistimed interventions, or inappropriate treatment recommendations, compromising patient safety.6,13,22,45
2. Health Inequities from Inequitable or Biased AI Tools
When AI tools are not intentionally designed and tested for fairness, they often generalize poorly to underrepresented populations.6,45 This can lead to unequal diagnostic accuracy, inappropriate treatment recommendations, and the amplification of existing health disparities.13,43 Rather than closing care gaps, biased AI may widen them, producing outcomes that contradict professional and ethical obligations to fairness, equity, and inclusion.13,46
3. Poor Generalizability in Real-World Clinical Settings
AI models developed and tested in controlled research environments may perform inconsistently when deployed in real-world orthodontic practice.22,45 Without demonstrated generalizability across patient populations, imaging conditions, and clinical workflows, these systems may yield unreliable or non-reproducible outputs, reducing clinical effectiveness and increasing the potential for patient harm.6,13
4. Reduced Access to AI Tools for Marginalized Populations
When bias is not proactively identified and addressed during development, AI tools may underperform for marginalized or underserved groups.22,45 As a result, these systems often cannot be safely deployed for those populations, effectively excluding them from access to AI-enabled diagnostics and treatment planning technologies.13,29,43 This lack of equitable design and validation perpetuates existing disparities in care, even as AI adoption expands.39
5. Clinical Risk from Unexplainable or Opaque AI Outputs
When AI models do not provide interpretable reasoning behind their outputs, orthodontists cannot meaningfully evaluate or challenge recommendations.47,48 This lack of explainability undermines clinical oversight, making it difficult to detect errors, assess appropriateness, or justify treatment decisions.6,13 Opaque models therefore pose direct risks to patients, since unsafe or inappropriate outputs may go undetected, leading to errors in diagnosis, treatment planning, or care delivery.22,43
6. Patient Harm from Unmonitored Performance Drift
Adaptive AI models are not static. Over time, they may degrade or behave unpredictably in response to changes in input data, clinical practices, or environmental conditions.6,10 Without continuous monitoring, validation, and drift detection, these performance shifts can go unnoticed and result in silent failures that erode clinical confidence and place patients at risk.19,24
7. Accumulation of Hidden Bias Over Time
Dynamic or continuously learning AI systems may acquire new biases after deployment if not proactively managed.6,45 Without routine fairness audits and demographic performance evaluations, these models can gradually develop patterns that disadvantage certain groups.24,43 This slow drift in fairness may go undetected, compounding inequities in care and undermining patient trust over time.13
8. Reinforcement of Outdated or Biased Clinical Norms
When AI models are trained on historical data that reflect outdated or biased clinical practices, such as overtreatment trends, racial disparities, or gender bias, they risk amplifying past harms rather than correcting them.45,49 Without intentional bias mitigation and dataset review, AI can codify and perpetuate these inequities under the appearance of “evidence-based care.”6,13
9. Undermining the Orthodontist’s Role as Human-in-Command
The orthodontist’s authority as the HIC relies on a baseline level of transparency and trustworthiness within AI tools.24,28 Without clear insight into how systems generate and update their outputs, orthodontists cannot effectively validate recommendations, exercise oversight, or fulfill their professional responsibility for patient care.13,21,22
10. Undisclosed Algorithm or Dataset Changes
When vendors modify algorithms or retrain models under a PCCP, those changes may be reported to regulators but not communicated to clinicians.19,22 Without clear and timely disclosure to end users, orthodontists may unknowingly rely on systems that no longer behave as originally validated or expected.6,24 This loss of transparency compromises the orthodontist’s ability to maintain oversight, verify appropriateness of use, and protect patient safety.6,19,22
11. Breakdown in Trust and Innovation Adoption
Untrustworthy AI systems do not just fail individually; they erode confidence in the broader use of AI within orthodontics and medicine.6,13 A single unethical or unsafe deployment can stall adoption, dissuade clinicians from integrating AI into practice, and set back innovation across the field.13,46
12. Legal and Ethical Exposure
Using unvalidated, biased, or opaque AI systems violates established ethical principles and exposes orthodontists, vendors, and institutions to significant legal risk, particularly when patient harm occurs and oversight of a licensed professional is insufficient.13,22,50,51
Call to Action
The AAO calls on developers to†:
a. To protect patients and ensure equitable care, developers must commit to building transparent, validated, and accountable AI tools from the start. The orthodontic community depends on vendors to uphold trustworthiness through ethical design, documentation, and continuous collaboration with clinicians.
2. Use Quality, Diverse, Clinically Verified Data
a. Train and validate models using authentic, high-fidelity clinical data – synthetic or low-quality datasets should only be used if equivalence is validated.20,22
b. Ensure broad demographic representation across age, ethnicity, gender, and geography to prevent performance disparities.13,43,45
c. Maintain dataset documentation that includes source provenance, labeling protocols, preprocessing steps, and version history to ensure auditability.6,28
d. Conduct regular bias audits and fairness testing throughout the AI lifecycle, reporting demographic performance differences and mitigation steps.6,13,43,45
3.Validate According to Risk Tier
a. Follow risk-based validation standards consistent with FDA and global guidance for SaMD, ensuring that the level of evidence is proportional to clinical risk.4,22,24,29,51
b. Use graded validation strategies proportional to risk, such as multi-site or randomized studies for high-risk AI models, and require independent external validation datasets that reflect the intended population.6,22,24,29
c. Publish key performance metrics (e.g., accuracy, sensitivity, specificity, F1-score) and detailed validation methodologies to enable peer and clinician evaluation.6,22,24
d. Include postmarket validation and re-verification schedules with continuous monitoring and drift detection to confirm ongoing safety and accuracy over time.6,19,24
4. Enable Full Explainability and Transparency
a. Integrate explainable AI features such as heatmaps, overlays, feature attribution, and confidence scores, so that clinicians can understand model reasoning and uncertainty.6,47,48
b. Provide clear, accessible documentation of intended use, known limitations, and interpretability safeguards to support safe clinical use and informed oversight.22,28,29
c. Ensure clinicians can meaningfully review, interpret, and override AI-generated outputs at the point of care, preserving HIC responsibility.13,22.47,52
d. Include traceability that links each output to model version, dataset lineage, and update history to enable audit and accountability across the lifecycle.6,22,24,28
5. Implement Continuous Monitoring and Update Protocols
a. Employ real-time drift detection, error tracking, and performance alerts accessible to both vendors and clinicians to maintain transparency and safety.6,19,24
b. Establish and follow clearly defined Algorithm Change Protocols (ACPs) and PCCPs to govern updates in adaptive AI models, documenting all retraining and validation processes.22,24
c. Provide postmarket performance dashboards to regulators and clinicians that report drift, retraining events, and update outcomes in an accessible and auditable format.6,19,28
d. Collaborate with orthodontists and professional organizations to integrate clinician feedback into model updates, ensuring real-world alignment, accountability, and shared oversight.6,28,29
6. Disclose Limitations and Maintain Documentation
a. Publicly release Post-Validation Reports (PVRs) summarizing data sources, validation results, and risk-mitigation strategies to promote transparency and accountability.22,24,28
b. Clearly disclose known limitations and failure conditions, identifying scenarios where the system may underperform or require manual oversight.13,22
c. Notify clinicians of any algorithm or dataset change that could affect model performance, not only those reported to regulators, to maintain real-world safety and trust.6,19,22,29
d. Maintain traceability logs, structured labeling, and accessible documentation in compliance with ISO/IEC 42001:2023 and related governance standards.6,24,28
e. Communicate proactively and clearly with clinicians, ensuring orthodontists have the information needed to uphold HIC oversight and patient safety.13,21
†For a detailed outline of these developer responsibilities, including standardized reporting templates and communication requirements, see the AAO’s forthcoming supplemental guidance on AI Transparency, Fairness, and Vendor Accountability.
The AAO calls on regulatory bodies to:
Protect the Public Trust
1. Regulators should ensure that all AI/ML SaMD tools used in clinical orthodontics are appropriately classified, registered, validated, and regulated under medical device standards consistent with FDA guidance and the IMDRF SaMD framework, with clear boundaries from CDSS exemptions.4,22,24,52
2. Regulators should expect developers to maintain regulator-reviewable documentation of dataset sources, demographic composition, labeling, and preprocessing, and to conduct fairness audits with stratified performance analysis during both premarket and postmarket phases.6,28,29,39
3. Regulators should require explainability and practical transparency for high-risk applications. Models are expected to provide interpretable outputs, such as confidence scores or visual saliency maps, sufficient for clinician oversight, consistent with the FDA AI/ML Action Plan and Good Machine Learning Practice principles.5,6,13,22,37
4. Regulators should support continuous oversight and real-world surveillance by promoting TPLC models that include real-time performance monitoring, drift detection, error reporting, and re-verification after updates.6,19,22,24
5. Regulators should establish clear accountability mechanisms for vendors whose tools underperform due to undisclosed updates, performance drift, bias, or inadequate validation. PCCPs and ACPs should include provisions for communicating clinically relevant updates and release notes to clinicians.19,22,50,51,53
6. Regulators should enforce risk-tiered clinical validation expectations by setting performance requirements appropriate to the device’s clinical risk classification, requiring independent external validation representative of the intended population, and preventing deployment absent sufficient evidence.6,20,22,24,29
7. Regulators should require lifecycle transparency and traceability by ensuring that clinician- and regulator-accessible documentation includes data sources, intended use, training methods, version history, update protocols, retraining triggers, and monitoring practices. Public summaries should be encouraged where feasible without breaching privacy or intellectual-property protections.6,22,24,28,29
The AAO calls on orthodontists to:
Govern with Oversight
As licensed healthcare professionals, orthodontists should remain the final decision-makers in all AI-assisted care. The licensed professional is expected to protect patient safety, uphold clinical integrity, and ensure the ethical implementation of AI/ML SaMD systems.13,21,22,24
1. Orthodontists should request documentation on dataset diversity, authenticity, and clinical validation before implementing any AI tool.22,29 Confirm that models were trained on representative, real-world orthodontic data encompassing varied ages, ethnicities, and geographies.22,43,45 Avoid tools built on low-fidelity or unverified data sources.6,45
2. Orthodontists should adopt only AI tools that demonstrate performance appropriate to their clinical-risk classification and provide evidence of independent, external validation before clinical use.22,24,29 Thresholds reported in the literature for high-risk diagnostic applications often exceed 90% for accuracy or similar metrics, but these figures are illustrative, not prescriptive. Regulators evaluate adequacy of performance evidence relative to risk classification, intended use, and validation context.22,24,29,41,54
3. Orthodontists should require sufficient explainability for every AI-generated recommendation to ensure safe and informed clinical use.13,22 Use only systems that provide interpretable outputs, such as confidence scores, visual saliency maps, or feature attribution tools, and confirm that these explanations are accessible and understandable to the clinician.6,48 Orthodontists should be able to interpret, question, and override results as the HIC, maintaining final accountability for clinical decisions.24,47,52
4. Orthodontists should review AI-assisted decisions periodically to detect performance drift or unsafe outputs and remain informed about software updates or retraining events that could affect clinical behavior.22,24 Any deviations or unexpected outcomes should be documented, and concerns reported to vendors or regulators to support continuous postmarket surveillance and system improvement.6,19
5. Orthodontists should request PVRs, update histories, and documentation of known system limitations to verify that AI tools remain transparent, validated, and regulatory-compliant.22,29 Clinicians should decline or discontinue use of tools that lack adequate transparency, fail to demonstrate safety or regulatory compliance, or do not provide sufficient evidence of validation.22,28 Providing structured feedback to vendors when performance or risk diverges from clinical expectations supports accountability and continuous improvement across the AI lifecycle.24,29
6. Orthodontists should participate in continuing education (CE) that focuses on the ethical use, explainability, and regulatory oversight of AI/ML SaMD to ensure competent and responsible implementation in clinical practice.1,6,13,41 All team members should understand system functions, limitations, and escalation protocols to maintain safe, informed integration of AI technologies into patient care.21,41
Core Principle 4: Patient Autonomy
Respect for patient autonomy requires full transparency about both the use of AI in clinical care and the use of patient data in AI development, validation, or retraining.13,39,45,55 Patients have the right to know when AI tools are involved in their diagnosis or treatment and when their data, whether identifiable or de-identified, contribute to AI system improvement.13,22,39,56 Clinicians and developers should ensure that data governance and auditability standards are in place to document dataset provenance, composition, and secondary use consistent with ethical and technical best practices.28
AI algorithms rely on large volumes of clinical data to learn, predict, and assist in decision-making. In orthodontics, these datasets are often privately owned and not subject to external review, raising concerns about data provenance, diversity, and consent for secondary use.45,55 The unregulated collection and reuse of patient data for model training can erode public trust, amplify bias, and compromise data integrity.13,56,57
Orthodontists should disclose when AI tools meaningfully contribute to patient care and ensure that patients are informed if their de-identified data are used to improve or retrain algorithms.13,30,39 Communication should be clear, concise, and patient-friendly, emphasizing the supportive, not substitutive, role of AI in clinical decision-making.13,58
Risks to Patient Care from Lack of Disclosure and Transparency
Failure to establish clear disclosure and transparency standards for the use of AI and patient data in orthodontic care may result in:
1. Erosion of patient trust and diminished public confidence in orthodontic providers and emerging technologies when AI use or data handling is not openly disclosed13,41,58
2. Legal and ethical exposure for clinicians if patients are unaware that AI contributed to their care or that their data were used for algorithm development or retraining.22,39,53
3. Compromised clinical accountability and oversight when AI-generated recommendations are applied without sufficient human verification or without disclosing their use to patients.24,29,30
4. Propagation of bias and inequity, as undisclosed algorithmic limitations and non-representative datasets may produce variable accuracy across patient populations, exacerbating healthcare disparities.43,45,55,57
5. Loss of patient autonomy and recourse, as lack of transparency limits patient understanding, informed participation, or the ability to question or refuse AI-supported recommendations.13,39,58
Call to Action
The AAO urges state dental boards and legislatures to take immediate action by recommending that:
1. Clinicians document in patient records when AI or AI-enabled tools meaningfully contribute to diagnosis, treatment planning, or care delivery, and when patient data, whether identifiable or de-identified, are used to support AI system development, validation, or retraining.
2. Communication standards should be established to ensure patients receive clear, patient-friendly explanations of how AI supports their care and how their data may be used, including the technology’s capabilities and limitations, and the orthodontist’s ongoing oversight.
3. State-level policies and guidance align with national and international AI governance frameworks to safeguard patient rights, ensure data privacy, and uphold ethical standards for transparency and accountability.
4. Education and professional training initiatives should be supported to equip dental professionals with the knowledge and skills to use AI responsibly, communicate transparently with patients, and uphold robust data governance and cybersecurity practices.
Core Principle 5: Education and Clinical AI Competency
As licensed healthcare professionals, orthodontists carry both an ethical obligation to patients and legal accountability for the technologies they integrate into care. Education in AI must therefore extend beyond basic literacy to encompass the knowledge and judgment required to evaluate, implement, and manage these tools responsibly. Orthodontists serve as the HIC at the point of care, exercising comprehensive oversight that includes selecting appropriate AI systems, verifying their clinical validity, ensuring their use aligns with licensure, and maintaining accountability for every AI-supported decision. Through structured education and deliberate competency development, orthodontists can ensure that AI strengthens professional judgment rather than replaces it.
Understanding how AI tools are developed, how their training data are sourced and validated, and how these models evolve through continuous learning is essential to protecting patient safety and professional integrity. Because AI systems are dynamic and may change over time, orthodontists must verify that each tool remains clinically validated, used within its intended context, and subject to ongoing human oversight.
Continuous, standardized education across all stages of training and professional practice is critical to fulfilling this responsibility. By advancing AI literacy throughout the profession, orthodontists can safeguard patient welfare, uphold clinical accountability, and lead in the responsible and ethical integration of emerging technologies.
AI Education Framework: Three-Tier Competency Model
1. Foundational AI Literacy
Orthodontists working directly or indirectly with AI tools should develop a basic understanding of AI principles, including ML fundamentals, data bias awareness, and AI’s role in orthodontics. To ensure responsible implementation, continuous human oversight, clinical agency, and ongoing validation are essential.1 There is a strong need to provide CE for clinicians, along with recommendations for orthodontic programs to incorporate AI literacy into their curricula. AI tools are increasingly being integrated into dental and orthodontic training programs, enriching professional development through simulations, diagnostic tools, and virtual patient models. These technologies enhance diagnostic and treatment planning capabilities while equipping clinicians with modern tools to improve care, efficiency, and precision.45,59 It is imperative that the end user understands the limitations of an AI system as well as who bears the responsibility in patient care.60
A core curriculum should outline essential knowledge that dental professionals must possess regarding AI. This curriculum must emphasize basic definitions, ML types, the role of training-validation-testing splits, and the concept of dynamic versus static AI systems. It should also underscore the importance of understanding AI explainability, reference tests, and the black box nature of most models—skills vital for building foundational AI literacy among clinicians.2 While AI applications show significant promise, manual supervision is still required to mitigate errors and variability. Continuous education for orthodontists is crucial, as it equips them with the ability to critically assess AI outputs, understand the underlying algorithms, and recognize their limitations. This ongoing education reinforces the importance of human expertise in validating AI-driven decisions and encourages the responsible use of these technologies in clinical practice.
Preclinical & Residency Programs
AI competencies should be integrated across undergraduate and postgraduate levels through a domain-based structure: understanding AI foundations, identifying dental-specific use cases, evaluating AI technologies, and addressing ethical or governance issues. This structure encourages the development of case-based learning, AI simulations, and outcome-based instruction, aligning well with the needs of orthodontic training programs aiming to blend AI into clinical learning environments..55
Integrate AI literacy, diagnostic simulations, and case-based virtual learning modules into orthodontic education to prepare future clinicians with foundational and applied AI competencies. Examples are as follows:
a. Simulation of Clinical Environments – AI-driven simulations can be used for training on orthodontic procedures and patient management.
b. Virtual Patients and Case-Based Learning – Implementation of virtual patients with AI-driven decision-making to simulate real-life cases .61 Generative AI tools can be used to enhance automated feedback systems, providing immediate, personalized feedback on residents’ clinical performance and treatment planning while fostering critical thinking and decision-making; and
c. Personalized Learning Paths AI-Powered Diagnostic Tools – Integration of AI in teaching students to interpret diagnostic data and its metrics.62
2. AI Training for Clinical Readiness
Orthodontists should receive hands-on training using AI-powered tools for cephalometric analysis, treatment simulations, appliance design, etc. Training should include interpreting confidence scores, identifying uncertainty, assessing bias, and understanding AI-generated outcomes.27,41 Integrating AI into dental education and training presents both opportunities and responsibilities. As Kim et al.63 emphasize, AI tools should augment clinical decision-making, not replace it. One key challenge is addressing inherent biases within AI algorithms, which may exacerbate health disparities if left unchecked.
Ethical AI education must help clinicians understand AI’s limitations, critically assess its outputs, and navigate concerns such as data privacy, security, and regulatory compliance. A stepwise integration approach is recommended, allowing orthodontists to build competency and confidence progressively as they adopt AI technologies into clinical practice.
Furthermore, it is essential to have hands-on experience with the typical architecture of AI tools, including backend algorithms, cloud-based inference systems, and model modularity. The curriculum also teaches learners how to evaluate AI systems using performance metrics such as accuracy, sensitivity, specificity, and Dice scores, enabling orthodontists to effectively assess AI outputs. These structured training elements equip clinicians to critically interpret diagnostic models and recognize how biases or low generalizability can impact treatment quality.55
3. Continuous Learning
Current and future applications of AI in orthodontics necessitate that licensed professionals continuously update their knowledge and skills. Ongoing education ensures that orthodontists can make informed, ethical decisions when selecting, validating, and supervising AI tools for clinical use. Continuous learning should highlight both the opportunities AI presents to improve accuracy, efficiency, and personalization of care, and the ethical challenges it introduces related to bias, accountability, and data governance.
The curriculum for AI education in orthodontics must be adaptable and evolve continuously, keeping pace with AI advancements and changing regulations. As part of their ongoing professional development, orthodontists should stay current on emerging concepts such as algorithmic explainability, risk-based regulation, and interdisciplinary oversight. This will enable them to responsibly adapt their clinical use of AI tools as these technologies progress.55
Risks to Patient Care from Lack of AI Training
Untrained clinicians may misunderstand AI’s performance metrics, apply models beyond their intended use, or assume unjustified confidence in AI-generated outcomes. Challenges related to AI use, such as lack of generalizability, dataset bias, and over-reliance on opaque black box systems, can all compromise patient safety if clinicians are not adequately trained to supervise AI outputs critically.55 The following are risks to patient care related to lack of clinicians’ AI training:
1. Misinterpretation of AI Outputs Leading to Treatment Errors
Orthodontists without proper training may misinterpret AI-generated recommendations, increasing the risk of diagnostic errors and poor treatment planning.10,13,18,41
2. Over-Reliance on AI Without Critical Evaluation
There is a risk that residents and clinicians may become overly reliant on AI and neglect to develop critical thinking and problem-solving skills.64 First, unquestioning acceptance of AI-generated outputs can lead to misdiagnosis or delayed diagnosis if errors or algorithmic biases go undetected. Second, complex or atypical cases often fall outside the scope of AI training datasets. Patients with rare conditions or unique variations may require adaptive reasoning and clinicians who have not developed strong problem-solving skills may struggle to manage such cases. Third, over-reliance on AI risks an erosion of clinical judgment and patient-centered care. Effective orthodontic treatment extends beyond technical outputs; it requires weighing psychosocial, ethical, and contextual factors. A clinician dependent on AI may undervalue these dimensions, reducing patient trust and satisfaction. Finally, a lack of critical evaluation skills may prevent clinicians from recognizing biases in AI systems. If training data are not representative of diverse patient populations, AI outputs may disproportionately misguide treatment in underserved groups, leading to inequities in care delivery.
3. Ethical and Regulatory Non-Compliance
Lack of training on AI ethics, transparency, and evolving regulations can lead to violations of scope-of-practice laws or device misuse, resulting in legal liability and disciplinary action.50,51,53 Patients may be exposed to unvalidated or improperly supervised AI tools, increasing the likelihood of harm from inappropriate treatment recommendations or device failures. Regulatory non-compliance can also undermine patient safety safeguards built into healthcare oversight systems, leaving patients vulnerable to unsafe technologies.
4. Erosion of Clinical Authority
If clinicians are not equipped to oversee AI tools, there is a risk that critical decision-making shifts from providers to software, thereby undermining the orthodontist’s central role in patient care.30,53,65 When AI assumes an unchecked role in decision-making, patients may receive care that is technically consistent with algorithmic outputs but misaligned with their individual needs, values, or preferences.
5. Delayed Adoption and Innovation Fatigue
Without targeted training and support, clinicians may feel overwhelmed or skeptical of AI, slowing adoption and widening gaps between clinical practice and innovation.18,30,64,66 Delayed adoption of validated AI tools can slow access to innovations that improve diagnostic accuracy, efficiency, and personalization of care. Conversely, clinician burnout from poorly supported implementation may reduce the quality of patient interactions and overall care delivery. Patients are therefore indirectly affected when clinicians lack the training and resources to adopt AI responsibly.
Call to Action: Implementing AI Training in Orthodontic Education
To prevent avoidable patient harm and professional liability, the orthodontic profession must act now to close the AI training gap.
The AAO calls for the following from orthodontists, vendors, and institutions:
1. Integrate AI coursework into preclinical and residency programs
To ensure that future orthodontists are equipped to interpret, supervise, and responsibly use AI in clinical care, AI coursework should be integrated into both preclinical and residency-level education. This integration must go beyond general exposure and include basic structured instruction in core concepts such as ML fundamentals, training-validation-testing models, performance metrics, and AI explainability.
Clinical modules should introduce practical applications of AI, including its supporting role in diagnostic simulations, cephalometric analysis, treatment planning, and appliance design. The curriculum should also emphasize ethical use, regulatory awareness, and the clinician’s ultimate accountability when employing AI-assisted tools.54,55,67
2. Establish AAO-led continuing education courses
To ensure consistent, safe, and ethical integration of AI in orthodontic practice, we call on the AAO to develop continuing education (CE) courses. These courses should offer CE that prioritizes clinical oversight, ethical application, and regulatory compliance. CE courses should clearly define the orthodontist’s role as the supervising authority over AI-assisted diagnostics and treatment planning particularly for tools categorized as SaMD and tools implemented in electronic health systems.
The CE should include training on critical topics such as interpreting AI performance metrics (e.g., sensitivity, specificity, Dice scores), managing uncertainty and bias in AI outputs, understanding FDA and IMDRF classifications of AI tools, and applying principles of human oversight and autonomy as outlined in international guidelines. This approach will not only standardize clinician readiness but will also support legal and ethical accountability.55,68
3. Stimulate vendor-sponsored training programs
Vendors who develop and distribute AI tools for orthodontic applications must play an active role in clinician education. We recommend requiring that all AI systems introduced into patient care be accompanied by structured, vendor-sponsored training.
This training should clearly explain the tool’s intended use, functional scope, data limitations, and necessary safeguards for safe implementation. It should also include practical demonstrations of how the system processes information, interprets clinical inputs, and generates outputs. By formalizing vendor responsibility in the education pipeline, we strengthen accountability, reduce misuse, and promote a culture of shared responsibility between developers and clinicians for patient outcomes.69
4. Promote Faculty Development in AI Education
To successfully integrate AI into dental education, faculty members must be provided with sufficient AI training. This is essential due to current studies indicating that a considerable number of dental educators possess limited AI literacy.
This lack of knowledge can obstruct the effective adoption of AI technologies in dental education. Therefore, comprehensive training programs that concentrate on AI applications in dentistry should be provided to faculty members by the school administrators. This will allow them to confidently incorporate AI technologies into their curricula, ultimately improving the standard of dental education.
Core Principle 6: Operational Integration and Data Privacy
Operational integration and data privacy are foundational to the responsible use of AI in orthodontics. Without secure, interoperable systems that protect patient information and function reliably within clinical workflows, even the most advanced AI tool cannot deliver safe or effective care. Fragmented data infrastructures, inconsistent interoperability standards, and variable vendor accountability create tangible risks to safety, efficiency, and public trust.6,7,22,28,29 As AI tools become increasingly embedded in diagnostic and treatment processes, their reliability and safety depend not only on algorithmic performance but also on the integrity of the data infrastructure, interoperability of connected systems, and clearly defined professional accountability for their use.6,14,22,28,29
Risks to Patient Care from Interoperability and Data Standardization Challenges
Seamless integration of AI tools into practice management software, imaging platforms, and electronic health records presents ongoing challenges that can jeopardize efficiency, accuracy, and patient safety if not properly addressed.
1. Lack of Data Standardization and Interoperability
Without consistent data standards, AI systems in orthodontics risk misreading or corrupting patient information as data move between platforms.6,28 Even minor mismatches in file formats or communication protocols can cause loss of critical health data or introduce diagnostic and treatment errors that compromise patient safety.21,70
2. Limited AI Adoption Due to Interoperability Barriers
Failure to establish consistent interoperability standards across AI systems and clinical platforms can hinder adoption, reducing efficiency and delaying access to advanced diagnostic and treatment technologies.6,21,22,28
3. Inequitable Access Due to Inadequate Interoperability
When interoperability standards are fragmented or inconsistently adopted, smaller or resource-limited orthodontic practices face disproportionate barriers to implementing AI technologies. These disparities slow adoption, limit access to advanced diagnostic and treatment tools, and risk widening gaps in care quality across patient populations.14,29
Data Privacy and Ethical AI Use
The responsible integration of AI clinical tools into orthodontic practice requires a consistent focus on safety, interoperability, and patient data protection across all stages of use.6,14,22,28,29
AI systems that manage patient data must comply with privacy and security regulations, maintaining encryption, access control, and secure storage to prevent unauthorized access or misuse.6,20,22 Ethical governance of these systems ensures that AI-supported recommendations preserve patient autonomy, transparency, and clinician oversight, reinforcing the orthodontist’s responsibility for protecting patient trust and information integrity.13,14,29
AI systems that handle patient data must comply with privacy and security regulations to prevent unauthorized access, misuse, or data breaches. Compliance with the U.S. Health Insurance Portability and Accountability Act (HIPAA) and the Health Information Technology for Economic and Clinical Health (HITECH) Act, together codified at 42 U.S.C. Chapter 6A, Subchapter XXVIII, and related provisions of Chapter 156, as well as with international and technical standards such as ISO/IEC 42001 and 27001, ensures that patient information is protected through secure design, encryption, and controlled access.6,20,28,29,71,72
Risks When Data Privacy Is Not Upheld in AI-Enabled Orthodontics
1. Patient Data Breaches and Unauthorized Access
Weak encryption, inadequate access controls, or insecure data sharing expose sensitive health records to cyberattacks, theft, and misuse, violating HIPAA and eroding patient trust.6,20,71,72
2. Corruption or Loss of Clinical Data
Poor governance or inconsistent data standards can cause information loss or corruption during storage or transfer, resulting in diagnostic errors, disrupted treatment continuity, and compromised patient safety.6,22,28
3. Unauthorized Secondary Use or Re-identification of Patient Data
Inadequate de-identification and lack of transparency allow vendors or AI developers to repurpose clinical data for algorithm training or commercial use without consent, breaching ethical and legal standards.5,13,14
4. Regulatory Non-Compliance and Legal Liability
Failure to meet HIPAA, HITECH, or ISO standards exposes orthodontists and vendors to regulatory fines, breach-notification obligations, and potential litigation for patient harm or data misuse.28,29,71,72
Call to Action: Strengthening Operational Integration and Data Privacy in Orthodontic AI
1. Adopt and enforce standardized data-exchange protocols such as HL7 FHIR (Health Level 7 Fast Healthcare Interoperability Resources) and DICOM (Digital Imaging and Communications in Medicine) across orthodontic systems to prevent data loss and diagnostic error.6,28
2. Mandate compliance with HIPAA, the HITECH Act, and ISO/IEC security frameworks to ensure robust encryption, access control, and breach-response capabilities.6,20,71,72
3.Require vendor transparency and data-use governance, including disclosure of how patient information is collected, stored, and used for algorithm development.5,13,14
4. Provide equitable support for practice modernization, offering technical guidance or funding mechanisms to help smaller practices meet interoperability and security standards.14,29
5.Institutionalize continuous education on AI data ethics and cybersecurity, ensuring clinicians and staff remain competent stewards of digital health information.13,14,28
Call to Action: Framework for Evaluating and Implementing AI Tools in Orthodontics‡
As AI continues to shape orthodontic diagnostics, treatment planning, and workflow design, orthodontists must adopt a structured, evidence-based approach to evaluation and implementation. A three-phase model consisting of Evaluate, Implement, and Monitor provides a foundation for the responsible integration of AI clinical tools in orthodontics.
Evaluate
Orthodontists should identify specific clinical needs, verify that AI tools are validated on diverse, high-quality datasets, and confirm that each system complies with applicable regulatory and technical standards, including FDA, IMDRF, ISO, HIPAA, and HITECH requirements. Careful assessment of workflow compatibility and vendor transparency ensures that AI adoption enhances precision, safety, and efficiency rather than introducing new risks.
Implement
Pilot AI tools on a limited number of clinical cases before full deployment, ensuring performance aligns with clinician benchmarks and established quality measures. Staff must be trained to validate AI outputs, recognize system limitations, and maintain transparency with patients regarding the role of AI in their care. Integration should align with state licensure requirements and uphold the orthodontist’s HIC responsibility.
Monitor and Educate Continuously
Clinicians must conduct regular audits to monitor accuracy, bias, and real-world reliability, while maintaining ongoing education in AI ethics, cybersecurity, and data governance. Collaboration with vendors and professional organizations ensures continuous improvement and regulatory compliance.‡A detailed version of this framework, including step-by-step guidance, case examples, and performance benchmarks, is available as a supplemental guide: Framework for Evaluating and Implementing AI/ML SaMD in Orthodontics