Legal Barriers to Implementing AI-Enabled Emergency Medical Services in Korea: Focus on the Medical Service Act, Emergency Medical Service Act, and Personal Information Protection Act Authors: Eun-jeong Lee, Yeon-woo Choi, Yeail Joo, Min Lee, Hyekyeong Kim, Chaemin Shin Published: 2026 ABSTRACT AbstractObjectivesIn prehospital and in-hospital emergency care, artificial intelligence (AI)-enabled emergency medical services (EMS) systems are reshaping clinical workflows through real-time triage, automated hospital assignment, telemedicine, and biometric monitoring. However, their implementation in Korea has revealed substantial legal and regulatory barriers. This study analyzed the structural incompatibilities between AI-enabled EMS and Korea’s current legal framework, with particular attention to the Medical Service Act, Emergency Medical Service Act, and Personal Information Protection Act.MethodsWe conducted a statutory review and case analysis of pilot projects, including the Chungbuk Smart EMS Project, the AI HOTline in Ilsan, and Yonsei’s AI-Ambulance model. We also performed a comparative assessment of foreign regulatory frameworks, including the US Health Insurance Portability and Accountability Act and the EU General Data Protection Regulation.ResultsKey barriers included the restriction of remote care to interphysician communication under Article 34 of the Medical Service Act, the lack of legal mandates for real-time hospital capacity sharing under the Emergency Medical Service Act, and limits on real-time sensitive-data processing caused by vague emergency clauses in the Personal Information Protection Act. These barriers constrain the scalability of AI-enabled EMS in Korea.ConclusionsTo support lawful and secure integration of AI-driven emergency care in Korea, we propose legislative revisions that establish emergency exceptions for remote AI-supported guidance, require hospital resource disclosure, and provide safe-harbor protections for healthcare professionals who rely on certified AI tools. A harmonized legal infrastructure is essential for realizing the potential of digital transformation in emergency care.Keywords: Emergency Medical Services, Artificial Intelligence, Telemedicine, Legislation, Health Policy Introduction Emergency medical services (EMS) are among the most time-critical components of national health systems. For stroke, ST-elevation myocardial infarction (STEMI), major trauma, and sepsis, effective triage, destination selection, and early initiation of evidence-based pathways require tightly coordinated decisions from dispatch through definitive treatment [1]. Advances in artificial intelligence (AI) can support each link in this chain. Current AI-enabled EMS systems can prioritize calls, assist severity assessment, recommend destination hospitals by matching patient needs with real-time capacity, and stream telemetry, imaging, and video for teleconsultation. These functions may shorten time to treatment and improve adherence to clinical pathways across regions [1–4]. Korea has piloted several of these functions. The Chungbuk Smart EMS initiative used information and communication technology–integrated prehospital data flows and hospital auto-assignment [2]. The National Health Insurance Service (NHIS) Ilsan AI HOTline/AI emergency network supported interfacility teleconsultation and information sharing [3]. A Yonsei-led AI-ambulance model enabled real-time streaming and AI-assisted support during transport [4]. Together, these pilots demonstrate technical feasibility and clinical utility in real-world settings. These pilots nonetheless revealed structural legal constraints that impede national scale-up. First, Article 34 of the Medical Service Act (MSA) frames telemedicine as clinician-to-clinician practice, placing emergency medical technician (EMT)-initiated, AI-assisted teleconsultation in a regulatory gray zone when physician input may be most urgent [5]. Second, the Emergency Medical Service Act (EMSA) assigns duties to emergency medical centers and national networks but does not require machine-readable, near-real-time disclosure of hospital resources. In the pilots, optional and inconsistent data feeds undermined AI auto-assignment performance and raised concerns about system equity [2,6]. Third, the Personal Information Protection Act (PIPA) remains consent centered and provides limited, ambiguous legal bases for time-critical processing of biometric streams, geolocation data, images, and automated profiling. As a result, institutions remain uncertain about live streaming, cross-facility sharing, and secondary data use, which may encourage conservative implementation [3,4,7]. International experience provides relevant points of comparison [8]. The EU General Data Protection Regulation (GDPR) recognizes the protection of vital interests, the provision of health care, and public-health interests as lawful grounds for emergency data processing. The US Health Insurance Portability and Accountability Act (HIPAA) permits use and disclosure for treatment and to prevent serious threats to health or safety. Although neither regime mandates real-time capacity reporting, both provide clearer pathways for emergency data handling and accountability than current Korean law, supporting the need for targeted reforms and certification-linked safe-harbor concepts [9,10]. Thus, AI-enabled EMS in Korea is technically ready for broader implementation but remains constrained by law. Without clear authorization for AI-assisted teleconsultation, standardized capacity feeds, and emergency data-use clauses, implementation will remain fragmented and discretionary [2–7]. In this context, and against the backdrop of global AI governance frameworks emphasizing transparency, continuous oversight, and privacy [11–13], we examined the MSA, EMSA, and PIPA in relation to AI-enabled EMS. We identified legal barriers and proposed a roadmap for safe, scalable implementation [2–7,9,10]. Methods Study Design and Scope We conducted a multisource normative legal and policy analysis with case-study synthesis. The unit of analysis was a statutory provision or subprovision and its related subordinate rules, which were mapped to operational failure types observed in Korean pilot projects [2–7,9,10]. The Technology–Organization–Environment (TOE) framework was used as a sensitizing heuristic rather than a formal analytic structure. The technology layer covered AI-enabled EMS capabilities, including automated triage, destination matching, biometric streaming, and teleconsultation. The organizational layer captured workflow frictions identified in the pilots, and the environmental layer captured constraints under the MSA, EMSA, and PIPA. Together, these layers guided the barrier mapping within a sociotechnical view of healthcare AI governance [2–7,9,10]. Legal Sources and Eligibility The primary sources were the official consolidated texts of the MSA, EMSA, and PIPA from the Korea Legislation Research Institute (KLRI). We included provisions addressing (1) teleconsultation and remote clinical advice, (2) hospital-resource disclosure and EMS information networks, and (3) emergency processing of sensitive data. Publicly available subordinate rules and guidelines were screened, and unrelated provisions were excluded [5–7]. Case Study Sources and Selection Three representative, publicly documented pilots of national relevance were purposively sampled: (1) Chungbuk Smart EMS, which involved information and communication technology–enabled auto-assignment and regional data sharing; (2) the NHIS Ilsan AI HOTline/AI emergency network, which involved interfacility telesupport and imaging/biosignal exchange; and (3) the Yonsei-led AI-ambulance model, which involved real-time streaming and in-transit AI support. Aims, AI functions, data types, legal friction points, and policy outcomes were extracted from peer-reviewed publications, official communications, and program reports [2–4]. Comparative Legal Review To contextualize the Korean statutes, we reviewed the GDPR and the HIPAA Privacy Rule from official repositories, including EUR-Lex and the US Department of Health and Human Services. These frameworks were used as references for emergency-processing bases, lawful health-data sharing for treatment, and safe-harbor-like constructs applicable to time-critical, AI-assisted workflows [9,10]. Operational Definitions and Analytic Questions Four analytic questions were prespecified. First, for remote healthcare authorization, does the law permit EMT-initiated, AI-assisted teleconsultation in emergencies, and under what controls? (MSA) [5]. Second, for hospital-resource disclosure, is there a mandatory, machine-readable standard for near-real-time publication of emergency department (ED) and intensive care unit (ICU) capacity, key devices, and specialist on-call coverage? (EMSA) [6]. Third, for emergency data processing, are consent exceptions and safeguards explicit for live biometrics, imaging, and automated profiling? (PIPA) [7]. Fourth, for accountability and safe harbors, do statutes delineate liability sharing and good-faith protections for certified medical AI used according to protocol? (cross-cutting) [3,7]. Data Extraction and Coding Article numbers, clause titles, and practical implications were extracted with a structured template. Excerpts were coded by theme and operational category, and pilot friction points were mapped to the most proximate statutory clauses. Classification discrepancies were resolved by consensus between the two primary reviewers, and final clause-level mappings were confirmed by a third expert [2–7]. Interrater reliability statistics, such as Cohen’s kappa, were not calculated. Consensus-based resolution is consistent with normative legal analysis, in which statutory interpretation requires substantive legal judgment; however, this approach is acknowledged as a limitation in Section IV-5. Pilot selection was purposive and was based on public documentation, national-level relevance, and functional diversity across AI-enabled EMS use cases. The resulting case-availability bias is also acknowledged in Section IV-5. This study used only publicly available statutory texts, official pilot reports, and peer-reviewed literature. It did not involve human participants, patient data, or personally identifiable information; therefore, institutional review board review was not required. Comparative Synthesis and Proposal Derivation For each theme, we developed matrices comparing Korean statutory constraints under the MSA, EMSA, and PIPA; emergency-use anchors in the GDPR and HIPAA; and operational needs derived from the pilot projects. Clause-level proposals and operational specifications were then formulated from these matrices [2–7,9,10]. Results Clause-level readings of the MSA, EMSA, and PIPA were mapped to the workflow failures observed in the three pilot projects and triangulated against GDPR and HIPAA comparators. This analysis produced a statute–workflow friction map (Figure 1), theme-structured barrier summaries (Table 1), a clause-level amendment set (Table 2), and cross-jurisdictional contrasts (Table 3) [2–7,9,10]. AI-enabled emergency medical services (EMS) workflow with legal friction points. The left column presents the operational sequence from dispatch to emergency department handover: emergency call (119) → AI triage and severity assessment → hospital auto-matching → AI ambulance monitoring → teleconsultation → emergency department admission and treatment. The arrows indicate progression through the emergency-care pathway. The right-hand callouts summarize the main legal constraints at each workflow step: 1) under the Personal Information Protection Act (PIPA), emergency data-consent requirements remain unclear, limiting real-time use of biometric and location data during dispatch; 2) under the Medical Service Act, AI-assisted medical guidance lacks clear authorization, creating uncertainty around AI-supported triage guidance; 3) under the EMSA, no binding mandate requires machine-readable, real-time emergency department, intensive care unit, or operating room capacity data for auto-matching; 4) no clear legal basis covers live ambulance telemetry or video streaming; 5) Article 34 limits teleconsultation to physician-to-physician communication, restricting EMT-initiated, AI-assisted teleconsultation; and 6) AI-assisted care is not clearly defined in liability terms, and no explicit safe-harbor framework assigns responsibility for certified medical AI. EMT: emergency medical technician, EMSA: Emergency Medical Service Act, AI: artificial intelligence, ER: emergency room. Table 1Summary of legal barriers identified in AI-enabled EMS pilot projects, aligned with the friction points in Figure 1Pilot projectRemote medical supportData-sharing issuesPrivacy/consent issuesChungbuk Smart EMSAI analysis was forwarded to hospitals, but its legality under the Medical Service Act remained unclear.Hospital resource sharing was optional, with no mandate under the Emergency Medical Service Act.Live biosignal transmission lacked a clear emergency-consent basis.Ilsan AI HOTlineTeleconsultation was initiated by emergency medical technicians, although the Medical Service Act limits remote care.Cloud-based hospital information system use was available, but real-time integration remained inconsistent.Data processing without explicit consent raised compliance concerns.Yonsei AI AmbulanceFull-duplex streaming from the ambulance was not clearly covered under current law.Interhospital system linkage for decision-making remained limited.Pseudonymization requirements for AI training were not clearly defined.EMS: emergency medical services, AI: artificial intelligence. Summary of legal barriers identified in AI-enabled EMS pilot projects, aligned with the friction points in Figure 1 EMS: emergency medical services, AI: artificial intelligence. Table 2Proposed legislative amendments and corresponding legal articlesLawCurrent barrierProposed amendmentMedical Service ActRemote healthcare is restricted to provider-to-provider communication.Permit AI-supported telemedicine between emergency medical technicians and physicians during emergencies.Emergency Medical Service ActHospitals are not obligated to share real-time resource information.Mandate real-time data disclosure for designated emergency medical centers.Personal Information Protection ActNo explicit exception permits real-time data processing during emergencies.Add an emergency exception clause for live biosignal processing.AI: artificial intelligence. Proposed legislative amendments and corresponding legal articles AI: artificial intelligence. Table 3Jurisdictional comparison with HIPAA and GDPRLegal/technical categoryKoreaUnited States (HIPAA)European Union (GDPR)Consent exemption for emergency usePossiblePossiblePossibleReal-time data sharingDependent on voluntary hospital cooperationPossiblePossibleAI decision-support recognitionNot explicitly authorizedPartialNot explicitly authorizedLiability protection for medical AINot explicitly authorizedPartialPartialCross-institution data integrationDependent on voluntary hospital cooperationPossiblePossibleHIPAA: Health Insurance Portability and Accountability Act, GDPR: General Data Protection Regulation, AI: artificial intelligence. Jurisdictional comparison with HIPAA and GDPR HIPAA: Health Insurance Portability and Accountability Act, GDPR: General Data Protection Regulation, AI: artificial intelligence. Medical Service Act (Article 34): Remote Healthcare Restrictions Article 34 restricts remote healthcare to medical professional–to–medical professional interactions, creating barriers to AI-assisted EMT-to-physician consultation during ambulance transport [5]. In the pilot projects, live streaming of vital signs and video to physicians was technically feasible but legally ambiguous, which led to conservative adoption [4–6] (Figure 1, Table 1). Emergency Medical Service Act: Real-Time Resource Disclosure Gaps The EMSA defines the roles of emergency centers and networks but does not require standardized, near-real-time disclosure of ED, ICU, or operating room capacity, device availability, or on-call specialist coverage [6]. In the Chungbuk pilot, optional participation produced incomplete data feeds that reduced the accuracy of AI auto-assignment [2,6] (Figure 1, Table 1). Personal Information Protection Act: Emergency-Use Limitations PIPA’s consent-centered rules and limited emergency carve-outs complicate continuous biometric streaming, rapid interfacility imaging exchange, and automated profiling in AI-enabled EMS [7]. In the NHIS Ilsan AI network, these requirements raised compliance concerns despite the system’s clinical benefits [3,4, 7] (Figure 1, Table 1). Ambiguity in Liability and Safe Harbor Neither the EMSA nor PIPA provides safe-harbor protection for certified medical AI systems used under established protocols. As a result, liability for AI-related errors remains uncertain [9,10] (Figure 1, Table 1). Clause-Level Amendment Map Table 2 summarizes the clause-level proposals: an MSA Article 34 emergency exception authorizing AI-assisted, EMT-led telemedicine [5]; an EMSA subregulation mandating machine-readable capacity feeds [6]; and a PIPA emergency-use legal basis with logging, archiving, and postevent notification [7]. The amendment set also incorporates a predetermined change control plan (PCCP), drawing on the US Food and Drug Administration’s final guidance for AI-enabled devices, to support legitimate iterative model updates within emergency workflows [5–7,11]. International Benchmarking Table 3 compares emergency data-processing bases, capacity criteria, telemedicine scope, and liability or safe-harbor approaches across Korea, the United States under HIPAA, and the European Union under the GDPR. Although neither foreign regime mandates real-time capacity disclosure, both provide clearer legal bases for time-critical data processing than current Korean law, underscoring the need to strengthen PIPA’s emergency-use provisions [9,10]. In parallel, the World Health Organization’s regulatory considerations for AI in health emphasize transparency, risk management, external validation, data quality, privacy, and stakeholder governance. The EU AI Act’s high-risk classification for medical AI systems provides an additional process-oriented benchmark that is consistent with the clause-level safeguards proposed in this study [12–14]. Proposed Operational Specifications for Capacity Provision Based on the capacity-gap analysis, we developed operational specifications for algorithmic destination matching. The minimum dataset should include near-real-time ED occupancy and queue length, including boarded patients; ICU and emergency-care-unit availability; operational status of critical devices, such as ventilators, high-flow oxygen, and extracorporeal membrane oxygenation; operational status of time-critical services, including computed tomography, magnetic resonance imaging, catheterization laboratories, and operating rooms; on-call specialist coverage; and diversion or bypass status. Core capacity and device metrics should refresh every 5–10 minutes, and diversion changes should be pushed within one minute. Data exchange must be secure and machine-readable, using mutual transport layer security or OAuth 2.0 authorization, short-lived tokens, and immutable audit logs, consistent with the HL7 FHIR Situational Awareness for Novel Epidemic Response Implementation Guide and National Emergency Medical Services Information System v3.5.0 [2,3,15,16]. Pre-Specified Evaluation Metrics for Scale-Up To support reproducible national scale-up and evaluation of clinical and operational performance, we prespecified outcome and process metrics aligned with the emergency-care workflow. These metrics include the following: 1) Acceptance latency: time from dispatch initiation to hospital admission. 1) Acceptance latency: time from dispatch initiation to hospital admission. 2) Destination-matching accuracy: concordance between the AI-recommended hospital and final clinical judgment. 2) Destination-matching accuracy: concordance between the AI-recommended hospital and final clinical judgment. 3) Transfer times: total duration before hospital arrival. 3) Transfer times: total duration before hospital arrival. 4) Diversion and transfer-out rates: measures of how efficiently patient load is distributed across facilities. 4) Diversion and transfer-out rates: measures of how efficiently patient load is distributed across facilities. 5) Door-to-intervention intervals: door-to-computed tomography, door-to-needle, and door-to-balloon times. 5) Door-to-intervention intervals: door-to-computed tomography, door-to-needle, and door-to-balloon times. 6) Equity indicators: admission delay stratified by hospital type, including public versus private and urban versus rural hospitals, and by region. 6) Equity indicators: admission delay stratified by hospital type, including public versus private and urban versus rural hospitals, and by region. These metrics provide direct instruments for evaluating specific reforms. A PIPA emergency-use basis should reduce acceptance latency and door-to-intervention intervals by removing consent-related transmission delays. EMSA-mandated capacity feeds should improve destination-matching accuracy and diversion or transfer-out rates. The proposed MSA Article 34 amendment should shorten door-to-needle and door-to-balloon times by enabling earlier physician input during transport. Equity indicators would provide ongoing surveillance for algorithmic bias across regions and facility types. Together, these measures form a monitoring framework for evidence-based regulatory review and iterative policy refinement at the national scale [2–7,9,10]. Discussion Key Interpretations Based on the Results Technical maturity does not appear to be the main barrier to AI-enabled EMS adoption in Korea. Instead, the current legal framework is narrow in its remote-care provisions under MSA Article 34, silent on real-time capacity disclosure under the EMSA, and ambiguous regarding emergency data processing under PIPA. These gaps are reflected across the workflow annotated in Figure 1 and summarized in Tables 1–3, despite successful pilots involving triage, destination matching, and remote support [2–7,9,10]. Implications by Result Theme 1) Remote healthcare authorization under MSA Article 34 The pilot projects confirm the feasibility and clinical utility of ambulance-initiated, AI-assisted teleconsultation, but Article 34’s clinician-to-clinician framing leaves EMT-initiated consultations legally uncertain [4,5]. A narrow emergency exception, limited to time-critical situations and conditioned on physician oversight, documentation, and human-in-the-loop controls, would reduce this legal friction [4,5] (Table 2). 2) Real-time capacity visibility under EMSA The EMSA does not require standardized, near-real-time disclosure of capacity, device readiness, or specialist availability. In the Chungbuk pilot, optional reporting yielded incomplete data and reduced auto-assignment accuracy [2,6]. Mandating a minimum dataset with defined update intervals, application programming interface standards, security requirements, and audit requirements would improve algorithmic performance and equity at the auto-matching stage [2,6] (Table 2). 3) PIPA emergency-use basis for lawful real-time processing PIPA’s consent-centered structure provides limited clarity for continuous streaming, imaging exchange, and automated profiling. The Ilsan experience shows how this ambiguity can encourage conservative practice despite clinical benefits [3,4,7]. Codifying an explicit emergency-use basis with logging, role-based access, limited retention, and postevent notification would legitimize real-time processing while preserving privacy protections [7] (Table 2). 4) Accountability and safe harbor for certified medical AI Current statutes do not explicitly allocate liability or provide good-faith protections for certified medical AI systems used according to approved protocols. This gap may foster risk-averse behavior. A certification-linked safe-harbor framework with auditable logs and postevent review would reduce defensive underuse while maintaining accountability [9,10]. 5) Human-in-the-loop requirements, AI audit obligations, and hallucination-risk mitigation AI error is a critical operational concern in time-critical emergency settings, where hallucinated outputs, overconfident triage recommendations, or erroneous hospital matching may have irreversible consequences. Any regulatory exception or safe-harbor provision should therefore be expressly conditioned on three safeguards. First, AI outputs must function as decision-support inputs rather than autonomous directives, with final clinical authority retained by a licensed physician or, where permitted under the proposed MSA amendment, a supervised EMT. Second, all AI outputs and corresponding clinician decisions should be recorded in immutable, timestamped audit logs that are accessible for regulatory review without compromising confidentiality. Third, adverse outcomes linked to AI-assisted decisions should trigger institutional postevent review by a designated clinical safety officer, with reporting to the relevant authority. This process should be modeled on pharmacovigilance but extended to AI-specific failure modes, including distributional shift and model hallucination. Embedding these safeguards within the proposed Ministry of Food and Drug Safety (MFDS)-linked certification framework would create a structured account-ability architecture that supports compliance and continuous clinical-safety improvement [9–11,17]. Policy and Implementation Guidance To operationalize these findings while preserving the reported benefits of AI-driven ED triage, including improved prioritization, shorter waits, and more efficient resource allocation [8], we propose the following measures. 1) EMSA-based technical notice on capacity feeds A national minimum dataset covering ED and ICU capacity, key device readiness, specialist on-call status, and diversion status should be mandated. The mandate should include standardized update intervals, application programming interface–level security, and audit requirements, thereby supporting equitable destination matching and AI auto-assignment accuracy [2,6,8]. 2) MSA Article 34 interpretive notice or narrow amendment AI-assisted EMT-to-physician teleconsultation should be formally recognized as a time-limited emergency practice subject to physician oversight, documentation, and human-in-the-loop controls. This recognition would enable lawful prehospital use of AI-supported triage and decision support [5,8]. 3) PIPA emergency-use legal basis with safeguards An explicit legal basis for time-critical data processing should be introduced under PIPA. This basis should incorporate role-based access, real-time logging, limited retention, and postevent notification, thereby legitimizing continuous streaming and automated triage while maintaining privacy protections and clinician trust [7,8]. 4) Certification-linked safe harbor for medical AI Good-faith protections should apply when certified medical AI systems are used according to approved protocols with auditable logs and incident reporting. Such protections could reduce defensive underuse while preserving accountability, consistent with international governance models [8–10]. 5) Phased implementation roadmap In the short term, defined as 0–12 months, the Ministry of Health and Welfare should issue interpretive administrative notices clarifying the permissible scope of practice under current MSA Article 34 and PIPA emergency provisions. It should also establish a regulatory-sandbox designation for controlled expansion of AI-assisted teleconsultation and real-time biometric streaming in selected regional EMS systems, with mandatory monitoring, incident reporting, and data collection. During this phase, formal stakeholder consultations should be initiated with policymakers, hospital emergency departments, EMTs, patient advocacy groups, and AI developers to identify trade-offs, including privacy versus operational timeliness and physician workload versus EMT autonomy. In the medium term, defined as 12–36 months, targeted amendments to MSA Article 34, the EMSA capacity provisions, and PIPA emergency-use clauses should be introduced based on sandbox evidence. These amendments should be piloted in designated emergency medical centers using the prespecified metrics described in Section III-8 to evaluate safety, performance, and equity. AI-tool certification should be aligned with the Ministry of Food and Drug Safety software-as-a-medical-device pathway described below. In the long term, defined as beyond 36 months, full rollout should rest on a permanent legal infrastructure that includes statutory amendments, Ministry of Food and Drug Safety–linked certification, standardized EMSA capacity feeds, and PIPA emergency-use safeguards. The equity indicators, door-to-intervention metrics, and AI audit logs described here should be embedded in the national EMS information system to support iterative improvement and regulatory adaptation. 6) Interface with Korea’s Ministry of Food and Drug Safety software-as-a-medical-device certification pathway The certification-linked safe harbor requires a clearly defined threshold. In Korea, that threshold should be anchored to the Ministry of Food and Drug Safety software-as-a-medical-device approval framework, which is aligned with International Medical Device Regulators Forum risk-categorization principles [18,19]. AI tools for EMS triage, destination matching, and teleconsultation would generally constitute high-risk software as a medical device, corresponding to International Medical Device Regulators Forum Class III–IV, given their life-critical and time-sensitive use. Safe-harbor protection should therefore apply only to tools that hold Ministry of Food and Drug Safety marketing approval, maintain a Ministry of Food and Drug Safety–recognized quality management system, and operate under a PCCP analogous to the US Food and Drug Administration’s final guidance [11]. This approach would tie legal protection to demonstrable clinical safety rather than mere deployment. The Ministry of Food and Drug Safety should also participate in stakeholder consultations so that the certification pathway evolves with the proposed amendments [9–11,18,19]. Equity, Quality, and Workforce Considerations Without standardized capacity feeds, AI routing may prioritize well-resourced hospitals because their complete and timely data make them more visible to assignment algorithms. This could increase AI-assigned volumes at digitally mature facilities and widen access gaps for hospitals with limited digital infrastructure. Matching algorithms should therefore incorporate safeguards such as baseline inclusion policies, benchmarking by facility type, and capacity-adjustment buffers. Clear teleconsultation authority could also shorten time to treatment for stroke, STEMI, and trauma by enabling earlier physician input in regions where emergency specialists are scarce. To support fair, safe, and trusted nationwide implementation, equity monitoring metrics, AI quality thresholds, and competency training for physicians and EMTs should be embedded in the system. Training should address legal risk, AI oversight, and documentation [2–7]. Strengths and Limitations This study’s strengths include its reliance on primary legal sources and traceable pilot documentation, with clause-to-workflow mapping that can be verified through Figure 1 and Tables 1–3 [2–7,9,10]. Its limitations reflect the nature of normative legal analysis, including evolving statutes, the risk of translation drift between Korean and English legal texts, and case-availability bias. These limitations were mitigated by cross-checking KLRI texts and triangulating across pilot projects and comparator regimes. Interrater reliability statistics were not calculated, and consensus-based coding introduces subjectivity that larger coding teams could address through structured reliability protocols. The three pilots were large, publicly documented initiatives and may not represent smaller regional programs. Notably, emergency physicians at the pilot hospitals already communicate with paramedics in real time, so the documented constraints may reflect institutional caution under statutory ambiguity rather than the absence of clinical capability. Real-time resource disclosure also involves substantial technical complexity beyond legal mandates: bed status must dynamically track occupied, reserved, pending-discharge, and planned-transfer states. Future research should therefore address these architectural requirements alongside the legal reforms proposed here. Footnotes Conflict of Interest No potential conflict of interest relevant to this article was reported.