IEC 62304 software architecture principles and existing design pattern standards covering modular, maintainable design.
Eight required secure design practices documented: trust boundaries, least privilege, attack surface reduction, economy of mechanism, proven patterns, secure defaults, and debug access controls.
Architecture documentation addressing all eight practices (5.3.2 a–h) with explicit tailoring rationale for any not applied — undocumented trust boundaries are a top finding.
Maps to
IEC 81001-5-1: §5.3.2 Secure design best practices
Requirement text
The manufacturer shall establish an activity (or activities) to identify, enforce, and maintain secure design best practices. The manufacturer shall document secure design best practices, which should include but are not limited to: a) documenting all trust boundaries as part of the design; b) least privilege; c) using proven secure software items/designs where possible; d) economy of mechanism (striving for simple designs); e) using secure design patterns; f) attack surface reduction; g) removing backdoors, debug access and debug information used during development, or documenting their presence and protecting them from unauthorized access; and h) protecting any remaining debug information from unauthorized access. The manufacturer shall define a security architecture as part of defense-in-depth, including these practices as appropriate.
Why this clause exists
Attack surface area is a direct measure of exploitability: every unnecessary service, exposed interface, shared credential, or over-privileged process is a potential entry point for an adversary. Secure design principles — least privilege, economy of mechanism, fail-secure defaults, complete mediation, separation of privilege — are not theoretical ideals but engineering constraints that reduce the attack surface at the design stage, when changes are cheapest. Organizations that defer security design review to the testing phase discover that attack surface problems are architectural: eliminating an unnecessary service exposed in the final architecture requires retesting; granting least privilege to a process that was architected with broad permissions requires code changes across subsystems. IEC 81001-5-1:2021 clause 5.3.2 requires these best practices to be applied during the design phase, not patched in during security testing. The clause covers the classic set: minimal attack surface, memory management safe practices, secure defaults, error state failure, minimal interfaces, trust boundary identification, least privilege, and economy of mechanism — each representing a class of vulnerability that emerges when the principle is not applied during design.
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 recognized it as Consensus Standard 13-122 (date of entry December 19, 2022). FDA's February 2026 cybersecurity guidance, Section V, lists IEC 81001-5-1 among the "possible frameworks to consider" for secure product development, and FDA's recognition record for the standard notes that conformance "may not satisfy all the cybersecurity requirements outlined in Section 524B of FD&C Act".
EU MDR harmonization was originally targeted for May 2024 but postponed to May 2028.
Common gaps (what we see in audits)
- Trust boundaries not documented in architecture — Software architecture documentation does not identify trust boundaries between components, networks, and users. Without documented trust boundaries, threat modeling cannot systematically identify where data crosses security perimeters and where controls are needed.