Safety

Modern vehicles increasingly depend on electronic systems, complex sensing, and artificial intelligence to perform safety-related functions. Ensuring their safety requires addressing both system malfunctions and limitations in how intended functions perceive and respond to real-world conditions. ISO 26262, ISO 21448, and ISO/PAS 8800 provide complementary approaches to managing these risks, supporting coordinated safety engineering across development, validation, and operation.

ISO 26262:2018 addresses hazards caused by malfunctioning safety-related electrical and electronic (E/E) systems and their interactions. It integrates safety activities into the vehicle lifecycle, helping manufacturers and suppliers define safety requirements, implement protective measures, and verify that systems achieve the required level of safety. Its focus is on risks arising from malfunctions, rather than the adequacy of a function’s nominal performance. 

ISO 21448:2022 complements ISO 26262 by addressing hazards arising from insufficient functional specifications or performance limitations, even when no fault covered by ISO 26262 has occurred. It also considers reasonably foreseeable misuse. Particularly relevant to driver assistance, automated driving, and emergency intervention systems, it guides design, verification, validation, and operational activities to identify hazardous scenarios and reduce unreasonable risk associated with sensing and decision-making limitations. 

ISO/PAS 8800:2024 adds an AI-specific safety perspective, focusing primarily on machine learning. It addresses vehicle-level risks associated with insufficient AI outputs, systematic errors, and random hardware errors affecting AI elements. It also considers interactions with external AI elements that can influence vehicle safety. Published as a Publicly Available Specification (PAS), it describes safety-related AI properties that support a convincing argument that unreasonable risk has been addressed.

These documents can be applied together within a coordinated safety process: ISO 26262 addresses malfunction-related risks; ISO 21448 addresses limitations of the intended functionality; and ISO/PAS 8800 adds AI-specific considerations that intersect with both. For example, in an automated emergency braking system, a controller malfunction falls within functional safety, inadequate perception under challenging conditions raises SOTIF concerns, and an ML model’s unreliable pedestrian classification calls for AI-specific safety analysis. This combined approach connects component-level assurance with the safety of the complete vehicle function.