Cybersecurity requirements for AI-enabled medical device software increasingly converge on the same core goals but remain unevenly integrated into regulatory and quality systems. A comparative analysis published in JMIR Medical Informatics examined regulatory and guidance approaches in Korea, the United States and the European Union across premarket review and postmarket surveillance. All three jurisdictions focus on protecting the confidentiality, integrity and availability of information and device functions throughout the product life cycle. However, cybersecurity is implemented through different regulatory architectures, while patient-safety vigilance and vulnerability-focused security processes often remain separate, limiting consistent oversight of risks that do not immediately cause patient harm.
Shared Goals, Different Regulatory Architectures
Across the three jurisdictions, premarket cybersecurity expectations centre on secure development, risk assessment and evidence that controls have been verified and validated. The main differences lie in how these expectations are structured and how explicitly regulators require life cycle measures such as threat modelling, software bills of materials, vulnerability management and update governance.
Korea’s Ministry of Food and Drug Safety places strong emphasis on documentation within ISO 14971-based risk management. Its guidance also addresses generative AI medical devices and recommends identifying threats linked to the technology used. The US Food and Drug Administration treats cybersecurity more directly as a quality-system and design-control activity spanning the total product life cycle. Its approach includes threat modelling and statutory requirements for certain connected devices.
Must Read: Health Cybersecurity Faces AI Threats and Recovery Gaps
In the European Union, cybersecurity sits within the Medical Device Regulation and In Vitro Diagnostic Regulation through general safety and performance requirements, supported by Medical Device Coordination Group guidance. Manufacturers also operate alongside broader requirements covering network and information security, artificial intelligence, cybersecurity and data protection. These additional obligations broaden the risk perspective beyond traditional device safety, but they are not readily integrated into device-centred quality management processes. Across all three systems, security risks that mainly affect data, service availability or system resilience can therefore sit outside safety frameworks when they do not clearly translate into patient harm.
Postmarket Oversight Shows the Greatest Gaps
Fragmentation is most apparent after devices reach the market. Medical device vigilance systems are largely organised around adverse events and serious incidents involving patient harm or threats to health. Cybersecurity practice, by contrast, also depends on vulnerability disclosures, exploit intelligence, abnormal system behaviour and near-miss events that may never cross a patient-harm threshold.
This separation creates parallel information flows. Device regulators mainly receive safety-oriented incident reports, while cybersecurity agencies and information-security channels handle broader vulnerability and breach data. In the European Union, serious incidents and field safety corrective actions fall under medical device rules, whereas cybersecurity incidents are handled at system or operational level through mechanisms including the Network and Information Security 2 Directive, national computer security incident response teams and the European Union Agency for Cybersecurity. Device-level cybersecurity enters vigilance when an event qualifies as a serious incident under the applicable medical device regulations.
The same structural issue appears across the jurisdictions despite differences in reporting rules. Separate pathways can delay links between a newly disclosed vulnerability and its potential clinical consequences, complicate mitigation when shared infrastructure or multiple manufacturers are affected and reduce systematic learning from security events that cause no immediate injury. Monitoring limited to device performance and patient-harm outcomes may also miss risks involving third-party software, communication interfaces, configuration changes or unusual data flows.
Quality Systems Could Link Safety and Cybersecurity
A proposed route towards more coherent oversight is to keep safety and cybersecurity risk management conceptually distinct while integrating their operational controls within the same quality management system. Within ISO 13485, design and development processes can address safety hazards alongside threats to confidentiality, integrity, availability and system resilience. Postmarket processes can then connect vulnerability monitoring, security incidents and performance degradation with corrective and preventive action and change control.
This model broadens formally managed risk beyond direct patient harm. A manufacturer’s risk processes could include both safety hazards under ISO 14971 and wider operational and information risks, with links between them when a cybersecurity event has potential safety consequences. Information security, AI governance and enterprise risk management can remain separate disciplines while using shared quality-system mechanisms such as design inputs, supplier qualification, configuration management, internal audit and management review.
The approach also extends surveillance across software updates, model retraining and changing threat conditions. Predetermined change control plans could incorporate defined cybersecurity assumptions and triggers for reassessment, while ISO 13485 design controls and improvement processes provide a framework for continuing monitoring. The model retains separate risk concepts while coordinating shared operational controls. At healthcare organisation level, multidisciplinary review involving clinical, biomedical engineering, information security and quality expertise is proposed as a way to connect security events with clinical change management and escalation decisions.
Cybersecurity governance for AI-enabled medical software is increasingly aligned on what needs protection but remains fragmented in how risks are managed and reported across the product life cycle. Safety-focused frameworks can leave vulnerabilities affecting data protection, service continuity and system resilience partly outside device-centred oversight when patient harm is not immediately evident. Integrating cybersecurity controls with established quality management processes could connect threat assessment, postmarket monitoring, incident response, corrective action and software change control while preserving the distinction between safety and security risk. The approach still requires empirical evaluation as regulatory guidance continues to evolve.
Source: JMIR Medical Informatics
Image Credit: iStock
References:
Jung S & Son K (2026) Regulatory Approaches to Cybersecurity Risk Management for AI-Enabled Medical Device Software in Korea, the United States, and the European Union: Comparative Document Analysis. JMIR Med Inform;14:e94846.
