Ethical,legal,and social implications of AI-enabled digital therapeutics: A comparative analysis of risk-based regulatory frameworks in the European Union and Japan

2.1. Study design and analytical approach

This study adopts a narrative review combined with comparative institutional analysis to examine how the ethical, legal, and social implications (ELSI) of therapeutic artificial intelligence are operationalized within digital health governance frameworks.^22,23 The analysis is explicitly governance-oriented, prioritizing regulatory design, institutional arrangements, and implementation capacity rather than clinical effectiveness or algorithmic performance. ²² This methodological choice reflects the article’s central concern: how ethical principles are translated into practice across different institutional settings.

Therapeutic AI sits at the intersection of technology, regulation, and care. Its governance spans multiple legal domains—including medical device regulation, data protection law, and AI-specific regulation—as well as multiple institutional levels ranging from supranational authorities to professional communities. ²⁶ Under such conditions, questions of regulatory coherence, normative trade-offs, and lifecycle oversight are not well suited to quantitative synthesis. Narrative review approaches are therefore especially useful in digital health policy research where interpretive analysis and contextual integration are required.^27–29

A narrative review is appropriate here because the field is both heterogeneous and fast moving. Relevant evidence is distributed across empirical studies, legal analyses, policy documents, and institutional reports, and these different source types must be read together to understand how governance works in practice.^27,28 In therapeutic AI, this heterogeneity is not incidental but constitutive of the subject itself, since regulatory development is shaped by interactions among law, professional norms, technological change, and institutional capacity.

Comparative institutional analysis complements this approach by enabling systematic comparison between governance systems with distinct regulatory traditions. ²⁶ Rather than treating legal frameworks as static texts, it examines how regulatory instruments interact with professional practices, administrative procedures, and institutional capabilities. The combined methodology therefore makes it possible to analyze not only what rules exist, but also how those rules function in practice as governance infrastructure.

Two jurisdictions—the European Union and Japan—were selected for comparative analysis because both face similar demographic pressures and digital health challenges while operating under distinct regulatory architectures. The EU represents a multi-level supranational governance system, whereas Japan represents a nationally centralized regulatory model. This contrast offers a useful basis for examining how ELSI principles become embedded in institutional governance structures.

2.2. Conceptual framework: Therapeutic AI and ELSI as governance infrastructure

The analytical framework is grounded in two core concepts: therapeutic artificial intelligence and ELSI as governance infrastructure.

Therapeutic AI is defined here as AI-enabled digital therapeutics and closely related digital health interventions designed to influence patient behavior directly for therapeutic or preventive purposes.^8,30 This category includes evidence-based AI-enabled digital therapeutics (DTx), adherence-support algorithms, AI-enabled mental health interventions, and adaptive coaching systems embedded in mobile or wearable platforms. ⁸ Its defining characteristics are continuous interaction, adaptive learning, and sustained behavioral influence beyond discrete clinical encounters.

Unlike diagnostic AI systems that generate episodic outputs for clinician interpretation, therapeutic AI systems operate directly in patients’ everyday environments and often do so without immediate professional mediation. Several application domains illustrate this pattern. Mental health technologies provide a particularly clear example, with AI-enabled chatbots delivering cognitive behavioral therapy or emotional support without direct human therapist involvement.^21,31 Chronic disease management systems employ adaptive algorithms to support medication adherence, lifestyle modification, and self-monitoring. Behavioral coaching platforms use persuasive design techniques and personalized recommendations to encourage preventive health behaviours.

ELSI is conceptualized in this study not simply as a set of ethical principles, but as governance infrastructure—that is, as institutionalized processes, roles, and mechanisms that translate ethical commitments into operational practice across the AI lifecycle. Within this framework, concepts such as explainability, accountability, and participation are understood as organizational and regulatory practices rather than merely technical features or abstract normative ideals.^4,7,32

This conceptualization builds on scholarship that highlights the gap between ethical principles and their practical implementation. Mittelstadt argues that principles alone cannot guarantee ethical AI because they must be institutionally embedded to function as effective constraints on behavior. ⁴ Mazzucato and Schaake similarly emphasize that governing AI in the public interest requires institutional capacity rather than ethical aspiration alone. ⁷ Building on these insights, this study treats ELSI as the institutional infrastructure through which ethical commitments acquire practical force in the governance of therapeutic AI.^{4,7,19,30,32,33}

2.3. Literature search strategy

A targeted literature search was conducted to identify scholarship at the intersection of digital health, AI-enabled DTx, governance, regulation, and ELSI. The search strategy was designed to balance breadth with analytical focus.

The primary database search was conducted in PubMed because of its extensive coverage of biomedical and health policy literature. Complementary searches were performed in Google Scholar to identify interdisciplinary research in law, political science, public policy, and science and technology studies, as well as relevant grey literature and policy documents.

Search terms were organized into three conceptual groups. Digital health terms included “digital health,” “digital therapeutics (DTx),” “digital medicine,” “mobile health,” “mHealth,” “eHealth,” “telemedicine,” “wearable technology,” and “software as a medical device.” Artificial intelligence terms included “artificial intelligence,” “AI,” “machine learning,” and “deep learning.” Governance-related terms included “ethics,” “ELSI,” “governance,” “regulation,” “policy,” “oversight,” “accountability,” and “transparency.”

The search string combined these three concept groups using Boolean operators as follows: (“digital health” OR “digital therapeutics” (DTx) OR “digital medicine” OR “mobile health” OR “mHealth” OR “eHealth” OR “telemedicine” OR “wearable technology” OR “software as a medical device”) AND (“artificial intelligence” OR “AI” OR “machine learning” OR “deep learning”) AND (“ethics” OR “ELSI” OR “governance” OR “regulation” OR “policy” OR “oversight” OR “accountability” OR “transparency”). Variations of this core search string were adapted to the syntax requirements of each database.

This search strategy was refined iteratively. An earlier exploratory search using broader field specifications yielded a larger but less focused set of records. The refined approach described above produced a more conceptually coherent dataset for governance-oriented analysis.

An initial exploratory search in PubMed without field restrictions yielded 1,246 records. To obtain a more conceptually focused dataset, the search was subsequently restricted to records in which the same Boolean search string appeared in the title or abstract fields using the PubMed [tiab] tag. This refinement resulted in 420 records, which were then subjected to title–abstract screening. After screening for relevance to governance, regulation, or ethical analysis, 87 publications were retained for full-text assessment. From this set, 34 core publications were selected for in-depth analysis because they addressed regulatory frameworks, institutional governance, or ELSI implementation in digital health and AI systems directly.^22,23,27

Inclusion criteria covered publications addressing governance, regulatory, policy, or ethical dimensions of AI in healthcare; comparative analyses of regulatory approaches across jurisdictions; conceptual contributions to ELSI in digital health; policy analyses of regulatory instruments such as the AI Act, MDR, GDPR, or PMD Act; and studies examining institutional implementation of digital health governance.^1,3,14–17 Studies were excluded if they focused exclusively on clinical outcomes, algorithmic development, or model performance without relevance to governance. Publications examining purely diagnostic AI were also excluded unless they offered transferable governance insights, such as analyses of liability allocation or lifecycle monitoring of high-risk AI systems.

The search was conducted in November 2024 and iteratively updated between December 2024 and January 2026 to incorporate recent analyses of the EU AI Act, digital therapeutics regulation, and Japanese governance frameworks.

2.4. Policy and legal document analysis

In parallel with the literature review, official legal and policy documents were analyzed as primary sources. Documents were identified through targeted searches of EUR-Lex, European Commission repositories, and official Japanese government and regulatory websites, including the Cabinet Office, the Ministry of Health, Labour and Welfare (MHLW), and the Pharmaceuticals and Medical Devices Agency (PMDA).

For the European Union, the analysis focused on the Artificial Intelligence Act (Regulation (EU) 2024/1689), including annexes defining high-risk AI systems and requirements concerning transparency, risk management, and human oversight. ³⁴ The Medical Device Regulation (Regulation (EU) 2017/745) was examined for provisions governing software as a medical device. ³⁵ The General Data Protection Regulation (Regulation (EU) 2016/679) was analyzed with particular attention to provisions on automated decision-making and algorithmic accountability. ³⁶ The European Health Data Space Regulation (Regulation (EU) 2025/327) was included in order to assess emerging frameworks enabling the secondary use of health data for AI development. ³⁷

For Japan, key documents included the Pharmaceuticals and Medical Devices Act (PMD Act), which governs software as a medical device, ³⁸ and the Act on the Protection of Personal Information (APPI). ³⁹ Administrative guidance and regulatory reports issued by MHLW and PMDA were analyzed to understand implementation practices, including initiatives such as the DASH framework for adaptive post-market modification of AI-enabled medical software.^40,41

All legal and policy documents were analyzed using a common template covering objectives, scope, risk categorization, oversight mechanisms, lifecycle governance provisions, and references to ELSI principles. This structured comparison made it possible to identify the regulatory philosophies and institutional mechanisms through which ethical principles are translated into operational governance practice.

Taken together, the documentary analysis was embedded within a comparative institutional strategy that examined regulatory philosophy, the institutional embedding of ELSI, and lifecycle governance capacity, while acknowledging the limitations of a qualitative, document-based study design.

3.1. The European Union: Layered governance with ex ante risk management

3.1.1. High-risk classification under the AI act and MDR

The European Union has established a layered governance architecture for therapeutic artificial intelligence by combining horizontal AI regulation with medical device law and health-data governance. Building on the conceptual governance framework developed in this study, Figure 1 first outlines how legal regulation, professional mediation, lifecycle monitoring, and participatory oversight jointly structure the governance infrastructure for therapeutic AI. Within this framework, Figure 2 illustrates how the European Union’s governance operates through the interaction of the AI Act, the Medical Device Regulation (MDR), the General Data Protection Regulation (GDPR), and the European Health Data Space (EHDS), together with implementation at Member State level.^{17,18,23,35–37}OPEN IN VIEWER

Figure 1.

Conceptual governance framework for therapeutic AI. This figure illustrates the conceptual argument developed in this study. AI-enabled digital therapeutics increasingly shape patient behaviour in everyday life, creating governance challenges that cannot be addressed by ethical AI principles alone. The framework proposes a conceptual shift from principle-based ethics to ELSI as governance infrastructure, encompassing legal regulation, professional mediation, lifecycle monitoring, and participatory oversight. The lower section illustrates how these governance dimensions are institutionalized differently in the European Union and Japan.

OPEN IN VIEWER

Figure 2.

This figure illustrates the institutional governance architectures for AI-enabled digital therapeutics in the European Union and Japan. In the European Union, governance is structured through layered horizontal regulatory frameworks, including the Artificial Intelligence Act, the General Data Protection Regulation (GDPR), and the European Health Data Space (EHDS), combined with sector-specific regulation under the Medical Device Regulation (MDR). These frameworks are implemented through national authorities, healthcare organisations, and professional actors within a precautionary, codified, risk-based regulatory model. In Japan, governance is primarily organised through sector-specific legislation and administrative practice centred on the Pharmaceuticals and Medical Devices Act (PMD Act), the Act on the Protection of Personal Information (APPI), and regulatory initiatives such as IDATEN and DASH for software as a medical device (SaMD). This system emphasises adaptive oversight, professional mediation, and post-market learning. The figure highlights how ethical, legal, and social implications (ELSI) operate as governance infrastructure through legal rules, institutional processes, professional judgement, and participatory mechanisms across both regulatory systems.

Within this framework, most therapeutic AI systems are likely to fall within the category of high-risk AI systems where they function as safety components of medical devices or are themselves regulated medical devices subject to conformity assessment.^17,18,35,36 In practice, this means that AI-enabled DTx and related software intended for therapeutic intervention, patient monitoring, or clinically relevant behavioural support are typically captured by a regulatory pathway that combines medical-device classification with AI-specific obligations.^17,18

Under the MDR, software intended to provide information used for diagnostic or therapeutic purposes is commonly classified as Class IIa or higher, depending on intended purpose and the severity of possible harm in the event of malfunction. ³⁶ Where such software also qualifies as an AI system under the AI Act, high-risk status triggers a set of legally binding obligations extending beyond conventional device safety requirements. ³⁴ These include lifecycle risk management, data governance, technical documentation, logging, human oversight, transparency for deployers, and requirements relating to accuracy, robustness, and cybersecurity.^34,36

This layered structure is especially important for therapeutic AI because these systems do not merely support isolated decisions; they may shape patient behaviour continuously over time. For this reason, AI-enabled mental health applications, DTx and adaptive behaviour-support tools are more likely to be treated as governance-sensitive technologies when their outputs materially affect treatment, patient risk profiles, or clinically significant behavioural outcomes.^8,9,17,18

3.2. Japan: Adaptive governance through sectoral regulation and professional mediation

3.3. Comparative findings: Divergent governance philosophies

The comparison between the European Union and Japan reveals two distinct modes of embedding ELSI into governance infrastructure for therapeutic AI. Figure 1 summarises the conceptual governance framework developed in this study, highlighting how legal regulation, professional mediation, lifecycle monitoring, and participatory oversight jointly constitute governance infrastructure for therapeutic AI. Therapeutic AI creates forms of sustained behavioural influence in everyday life that cannot be governed adequately through abstract ethical principles alone, so effective governance requires institutional arrangements capable of addressing behavioural influence, lifecycle change, and accountability across AI-enabled therapeutic systems. Figure 2 then maps how this governance infrastructure is instantiated in the European Union and Japan, showing the layered supranational framework in the EU and the sectoral, practice-centred model in Japan. Together, these figures clarify both the conceptual argument and the institutional pathways through which ELSI is operationalised in the two jurisdictions.

Figure 1 highlights the central analytical move of the article: therapeutic AI creates forms of sustained behavioral influence in everyday life that cannot be governed adequately through abstract ethical principles alone. Instead, governance requires institutional infrastructure comprising legal regulation, professional mediation, lifecycle monitoring, and participatory oversight. The comparative implication is that effective governance of therapeutic AI requires a hybrid model combining legal safeguards with adaptive lifecycle oversight.

Figure 2 then shows how this general conceptual framework is instantiated differently across jurisdictions. In the EU, ELSI is embedded through a precautionary, codified, and risk-based model combining the AI Act, MDR, GDPR, and EHDS with implementation through Member State institutions and healthcare organizations.^{17,18,23,35–37} In Japan, ELSI is embedded through a more adaptive, sectoral, and practice-centered model grounded in the PMD Act, APPI, PMDA review practice, and professional mediation.^{24,25,33,38–41}

These different approaches can be summarized across seven dimensions: level of governance, legal basis, risk classification, lifecycle control, degree of ELSI formalization, the role of professionals, and innovation trajectory (Table 1).

OPEN IN VIEWER

Table 1.

High-risk AI regime for therapeutic AI systems under the EU AI Act and MDR.

Dimension	European union framework	Key references
High-risk AI trigger	AI systems used as safety components of MDR-regulated medical devices, or medical devices themselves where third-party conformity assessment is required	AI Act Art. 6, Annex III; MDR Annex VIII35,36
Typical scope	Therapeutic intervention, patient monitoring, AI-enabled DTx, and mental health AI where outputs materially affect clinical decisions, patient risk, or therapeutic outcomes	8,17,18
MDR risk class	Software for diagnosis or treatment is typically classified as Class IIa or higher; Class III may apply where malfunction could cause death or irreversible deterioration	36
Provider obligations	Lifecycle risk management, data governance, technical documentation, logging, deployer information, human oversight, accuracy, robustness, and cybersecurity	AI Act Arts. 9–15, Annex IV 35

OPEN IN VIEWER

Table 2.

Governance architectures for therapeutic AI in the European Union and Japan.

Dimension	European union (EU)	Japan
Level of governance	Supranational framework regulation (AI Act, MDR, GDPR, EHDS) layered with Member State implementation pathways17,18,23,34–37,45	Single-nation legal system centred on the PMD Act and APPI, complemented by ministry guidance, PMDA review practice, and professional mediation24,25,33,38–41
Primary legal basis for AI in medicine	Horizontal AI Act combined with MDR for medical devices and GDPR/EHDS for data protection and secondary use of health data17,18,35–37,45	No horizontal AI statute for medicine; AI-enabled systems governed mainly through PMD Act, APPI, and sector-specific implementation rules24,25,38,41
Risk classification of AI-enabled SaMD	AI-enabled medical devices typically classified under MDR and, where applicable, treated as high-risk AI systems under the AI Act17,18,34,35	AI-based SaMD classified through device classes under the PMD Act, with risk managed through device class, review route, and post-market controls24,33,38,40,41
Lifecycle and post-market governance	Legally binding lifecycle risk management, data governance, documentation, conformity assessment, and post-market surveillance34,35	Iterative updates and post-market modification supported by PMDA/MHLW review practice, including DASH-related adaptive approaches24,33,40,41
ELSI formalization	Fundamental-rights and ethical principles embedded as binding requirements for high-risk AI systems16–18,34	ELSI mainly articulated through sectoral regulation, administrative guidance, and professional ethics rather than through a single binding AI statute24,25,33,40
Role of professionals and deployers	Human oversight required, though professional roles are mediated through sectoral law and implementation practice16–18,23,34	Physician and institutional responsibility remain central in diagnosis, treatment, and deployment of AI-enabled systems24,25,41
Innovation and compliance trajectory	Higher ex ante compliance burden, but greater legal certainty and harmonization across markets4,7,17,18,23	Greater flexibility for iterative innovation, but potentially more variability in implementation and visibility of ELSI safeguards24,25,33,40,41

4.1. Beyond principle-based ethics: Therapeutic AI as a governance problem

The analysis suggests that the governance challenges posed by therapeutic artificial intelligence cannot be addressed adequately through principle-based ethics alone. Although ethical principles such as transparency, fairness, accountability, and human oversight have become central reference points in global AI governance discourse, these principles by themselves do not ensure responsible implementation in practice.^1–7,16,20 The key issue is not simply whether such principles are normatively desirable, but how they are translated into institutionalized processes operating across the lifecycle of AI-enabled interventions.

This matters because therapeutic AI differs in important respects from earlier forms of medical AI. Rather than generating discrete outputs for clinician interpretation, therapeutic AI systems may operate continuously in patients’ everyday environments, shaping behaviour through repeated interaction, nudging, adaptation, and longitudinal engagement.^8,9,13,14,21 Governance problems therefore emerge not only at the moment of system approval or deployment, but throughout the lifecycle of the system. Concerns related to autonomy, persuasion, opacity, and accountability may develop gradually through cumulative behavioral influence rather than through a single identifiable decision point.^16,19–21

The comparative results reinforce the central conceptual claim of this article: ethical, legal, and social implications (ELSI) should be understood not merely as ethical concerns, but as governance infrastructure. Ethical commitments become meaningful only when embedded in legal rules, institutional procedures, professional responsibilities, and mechanisms for monitoring and accountability.^{4,7,20,32,45–47}

4.2. The European Union: Codified and precautionary governance

The European Union’s governance architecture illustrates the strengths of a codified and precautionary model of AI regulation. By combining the Artificial Intelligence Act with the Medical Device Regulation (MDR), the General Data Protection Regulation (GDPR), and the emerging European Health Data Space (EHDS), the EU has constructed a layered regulatory framework through which therapeutic AI systems can be governed across multiple legal domains.^{17,18,23,35–37}

Within this framework, many therapeutic AI systems fall into the category of high-risk AI systems where they function as safety components of medical devices or are themselves regulated medical devices requiring conformity assessment.^17,18,34,35 In practice, this means that AI-enabled DTx and related systems are subject to a range of legally binding obligations, including lifecycle risk management, data governance requirements, documentation obligations, human oversight mechanisms, transparency requirements for deployers, and standards concerning robustness and cybersecurity.^34,35

One advantage of this approach is the creation of clear ex ante obligations. Developers and deployers can identify regulatory requirements before placing systems on the market, improving legal certainty and regulatory transparency.^17,18,34 In addition, by embedding ethical concerns into legally enforceable duties, the EU framework translates abstract ethical principles into operational governance mechanisms.^4,16–18

The EU model also has limits. The interaction of multiple legal instruments may create regulatory complexity, particularly for smaller innovators and for AI systems that evolve after deployment. Therapeutic AI systems may change through retraining, updates, or shifting patterns of real-world use, raising questions about how conformity assessment and lifecycle oversight should adapt to dynamic systems.^17,18,34 The EU approach therefore provides strong formal safeguards, but may encounter difficulties when regulating continuously learning systems.

4.3. Japan: Adaptive governance through sectoral regulation and professional mediation

Japan represents a contrasting governance model centered on sectoral regulation and administrative adaptation. Rather than creating a horizontal AI-specific legal framework, Japan regulates therapeutic AI mainly through the Pharmaceuticals and Medical Devices Act (PMD Act), data protection legislation such as the Act on the Protection of Personal Information (APPI), and regulatory practice developed by the Pharmaceuticals and Medical Devices Agency (PMDA).^{24,25,33,38–41}

Under this system, AI-enabled medical software is regulated as software as a medical device within the established medical-device classification framework. Risk classification and approval pathways are determined primarily by intended use and potential patient harm, rather than by the existence of a separate AI regulatory regime.^24,39 This approach embeds therapeutic AI governance within existing medical-device oversight structures and clinical regulatory processes.

A notable feature of the Japanese approach is its emphasis on adaptive post-market governance. Regulatory initiatives such as DASH-related frameworks enable structured management of updates and modifications for AI-based medical software, reflecting a regulatory philosophy that prioritizes real-world evidence and iterative improvement.^33,40,44 This allows oversight to evolve alongside technological development.

Professional mediation also plays a central role. Physicians, hospitals, and expert committees contribute to interpreting and applying regulatory frameworks in clinical contexts. Governance is therefore implemented not only through statutory obligations, but also through institutional practice and professional responsibility.^24,25,44

This approach has its own limitations. Because ethical safeguards are less explicitly codified in law, governance outcomes may depend more heavily on administrative interpretation and institutional practice. This may lead to greater variability in how ELSI principles are operationalized across different contexts.^24,25

4.4. Divergent governance philosophies

Comparison of the European Union and Japan highlights two different ways of embedding ELSI within governance systems for therapeutic AI.

The EU model is characterized by precautionary, codified, and risk-based governance, in which ethical principles are translated into legally binding requirements associated with high-risk classification and lifecycle compliance obligations.^{16–18,35–37} Japan’s approach, by contrast, is more adaptive, sectoral, and practice-centered, relying more heavily on administrative guidance, institutional oversight, and professional mediation.^{24,25,33,38–41}

These differences should not be reduced to simple variations in regulatory strictness. They instead reflect distinct strategies for assembling governance capacity. The EU emphasizes formal legal structures and harmonized compliance obligations, whereas Japan emphasizes flexibility, incremental learning, and sector-specific expertise.

The comparison suggests that neither model alone fully resolves the governance challenges posed by therapeutic AI. Highly codified frameworks may struggle to accommodate rapidly evolving technologies, while purely adaptive approaches may lack transparency and consistency. Effective governance may therefore require a hybrid model combining legal safeguards with adaptive lifecycle oversight mechanisms.

4.5. Implications for digital health governance

Several implications follow from these findings.

First, therapeutic AI should be governed as a category of technology that raises not only safety and performance questions, but also questions of behavioral influence and relational autonomy.^8,9,19,21,42 Governance frameworks designed for static medical devices may not be sufficient where digital systems shape patient behavior continuously in everyday settings.

Second, lifecycle governance should be treated as a central component of AI regulation. Therapeutic AI systems may evolve after deployment through software updates, learning processes, or changing patterns of use. This makes post-market monitoring, revision pathways, and feedback loops essential components of governance infrastructure.^{17,18,33–35,40,41,44,45}

Third, professional mediation remains important, but is not sufficient on its own. Clinicians and healthcare organizations play a critical role in supervising the use of therapeutic AI, but governance frameworks must also address situations in which AI systems operate with limited professional oversight, particularly in consumer-facing digital health applications.^19,21,24,25

Fourth, participatory oversight mechanisms deserve greater attention. If therapeutic AI systems influence behavior in everyday life, governance frameworks should incorporate mechanisms that allow patients and publics to contribute to oversight, evaluation, and accountability processes.^{16,32,44,45,47}

For developers, these findings imply that governance considerations need to be integrated into system design from the outset, including documentation, transparency mechanisms, update management, and monitoring capabilities. For healthcare organizations, they highlight the need for institutional governance capacity capable of evaluating, deploying, and supervising AI-enabled therapeutic systems.

4.6. Limitations and future research

This study has several limitations. First, the analysis is based on a narrative review and comparative institutional analysis rather than on a systematic review or empirical case studies. Its aim is therefore interpretive and conceptual rather than exhaustive or statistically generalizable.^27–29

Second, the analysis focuses primarily on formal governance frameworks and institutional design rather than on implementation outcomes in clinical practice. Empirical research examining how therapeutic AI governance operates within healthcare organizations and real-world deployment settings remains necessary.

Third, the comparison is limited to two jurisdictions. Future research should extend comparative inquiry to additional regulatory environments, including the United States and emerging Asian digital health governance systems.

Further empirical work is also needed to examine how patients, clinicians, regulators, and developers experience governance mechanisms in practice, particularly where AI-enabled therapeutic systems operate across hybrid clinical and consumer environments.