IEC 62304 §5.3 software architectural design and software design documentation covering algorithm and module-level design decisions.
Secure design at detailed level with threat-to-design-control traceability, security measures addressing each threat model entry, and technology-specific secure design considerations.
Software design document with explicit threat model linkage per design element — designs that address architecture security but skip component-level controls are a moderate finding.
Maps to
IEC 81001-5-1: §5.4.1 Software design best practices
Requirement text
The manufacturer shall establish an activity (or activities) to develop and document a secure health software design and maintain the use of best practices for the secure design, taking into account: a) software technology at application level (e.g., algorithms, methods); b) the programming technology used (e.g., programming language); and c) the secure design best practices in 5.3.2. The health software design shall include measures to address the threats identified in the threat model.
Why this clause exists
Design specifications that do not include security properties — that describe what a component does but not how it enforces access control, handles untrusted input, manages keys, or logs security events — produce implementations that achieve functional requirements while ignoring security requirements. Developers working from a functional specification without security detail will make implementation-level security decisions ad hoc, inconsistently, and without the benefit of a security design review that would have caught errors. IEC 81001-5-1:2021 clause 5.4.1 requires that software design specifications incorporate security design properties at the level of detail needed to implement and verify them. This means security is not a separately documented activity that runs in parallel — it is incorporated into the design specification that drives implementation, tested against the verification criteria derived from security requirements, and traceable through the requirements traceability matrix from threat to control to test. The result is that implementation-level security decisions are design-reviewed rather than developer-discretionary.
What changed
IEC 81001-5-1:2021 is the first standalone cybersecurity standard purpose-built for health software and medical device software. Published in December 2021, it was adapted from IEC 62443-4-1 (industrial control systems security) to address the unique safety and regulatory context of medical devices — adding health-specific requirements that account for patient safety, clinical workflows, and the manufacturer-HDO relationship.
The standard mirrors IEC 62304's lifecycle structure but adds security-specific activities at every phase — planning, development, testing, release, and maintenance. It requires security risk management to be integrated with ISO 14971 safety risk management, not treated as a separate IT concern. FDA formally recognized it as Consensus Standard 13-122 on December 19, 2022 and references it as providing one acceptable framework for satisfying the cybersecurity requirements of Section 524B(b)(2), which requires manufacturers to design, develop, and maintain processes and procedures to provide a reasonable assurance that cyber devices and related systems are cybersecure.
EU MDR harmonization was originally targeted for May 2024 but postponed to May 2028. Despite this delay, Notified Bodies and Competent Authorities universally recognize it as "state of the art" for health software cybersecurity under MDR GSPR Annex I, Section 17.2. Missing or inadequate cybersecurity documentation is already a top cause of Notified Body major non-conformities for SaMD. A December 2025 Interpretation Sheet (ISH1:2025) clarified software item classification into maintained, supported, and required software categories, affecting risk transfer and post-market obligations.
Common gaps (what we see in audits)
- Secure design principles not applied at detailed design level — While architectural security may be addressed, detailed design decisions (data validation, session management, error handling, logging) do not consistently apply secure design principles like least privilege, fail-safe defaults, and complete mediation.