本文章目前仅提供英文版本。
ISO/PAS 8800: The Safety of AI in Road Vehicles
Artificial intelligence (AI) is rapidly becoming a core element of modern vehicle functions - particularly in Automated Driving Systems (ADS). While existing automotive safety standards address many traditional risks, they were not designed to fully cover the data‑dependent and non‑deterministic behavior of AI systems. To address this gap, the automotive industry introduced ISO/PAS 8800:2024 "Road Vehicles - Safety and Artificial Intelligence", a publicly available specification that defines how AI‑related safety can be systematically managed within automotive development.
What ISO/PAS 8800 Does
ISO/PAS 8800 applies to safety‑related vehicle systems that include AI elements, with a strong focus on machine‑learning‑based components. Its objective is not to replace existing safety standards, but to extend them where AI introduces new sources of risk.
ISO/PAS 8800 introduces:
- an AI‑specific safety lifecycle aligned with automotive processes,
- concepts and terminology tailored to AI systems, and
- mechanisms to build assurance arguments demonstrating the absence of unreasonable risk due to AI behavior.
A defining characteristic of ISO/PAS 8800 is the recognition that data itself becomes a safety artifact. Training, validation, and test datasets must be specified, justified, and controlled as part of the safety lifecycle.
Relationship to ISO 26262 (Functional Safety)
ISO 26262 addresses hazards caused by malfunctions, such as systematic and random faults/failures. When AI elements are involved, many of these concepts remain applicable but require an interpretation adapted to AI architectures. ISO/PAS 8800 builds on ISO 26262 by:
- tailoring applicable clauses to AI elements,
- extending development and verification activities for AI‑related failures, and
- integrating AI behavior into safety cases/arguments where malfunctions are relevant.
Relationship to ISO 21448 (SOTIF)
ISO 21448, commonly known as SOTIF, addresses hazards that occur without faults, caused by functional insufficiencies. In AI‑based systems, SOTIF risks often stem from:
- perception limitations,
- incomplete specifications, or
- insufficient coverage of real‑world scenarios in training data.
ISO/PAS 8800 refines the SOTIF approach by drilling down to the level of AI components. It enables teams to identify AI‑specific functional insufficiencies, trace them to root causes such as data gaps or AI model limitations, and link AI‑level mitigations back to system‑level SOTIF measures.
A Complementary Safety Landscape
Together, the three standards address different yet complementary sources of risk:
- ISO 26262 (Functional safety) – hazards due to malfunctioning behavior, i.e. faults/failures
- ISO 21448 (SOTIF) – hazards due to functional insufficiencies without faults
- ISO/PAS 8800 (AI safety) – AI‑specific risks driven by uncertainty, learning behavior, and data dependence
This layered approach enables a more complete and defensible safety case for AI‑enabled vehicle functions.
Why ISO/PAS 8800 Matters for Automotive Engineers and Safety Managers
For engineers and safety managers working with AI‑enabled systems, ISO/PAS 8800 provides the missing link between established automotive safety standards and the unique challenges introduced by machine learning (ML). Understanding how it interacts with ISO 26262 and SOTIF is becoming essential for credible safety arguments in modern vehicle development.
How tudoor academy Can Help You With ISO PAS 8800
With direct involvement in the ISO/PAS 8800 standardization activities, tudoor academy experts from samoconsult combine safety standards knowledge with practical implementation experience. Through advisory services and focused training, we help teams build a solid understanding of AI safety in line with ISO/PAS 8800. If you would like to learn more about our ISO/PAS 8800 training or advisory offerings, please just contact us.
