ISO/PAS 8800: Training & Workshops, Audits and Assessments

Whether more safety on the road, personalized driving experiences or more efficient processes – these advancements in modern vehicles are virtually unthinkable without artificial intelligence. Huge amounts of data are generated by sensors, cameras and other systems, which then must be processed by machine learning and be analyzed using data analytics. Furthermore, advanced driver assistance systems and autonomous driving require complex sensor systems to make precise decisions in real time.

The challenge is to manage this flood of data while ensuring that safety-critical applications work correctly. This is where ISO/PAS 8800 comes in.

Published in December 2024, "ISO/PAS 8800 – Road Vehicles – Safety and Artificial Intelligence" is a safety standard for AI-based systems in vehicles. It defines a framework for the use of AI on electrical and electronic systems in safety-critical applications, and addresses risks that can result from malfunctions or inadequate performance of artificial intelligence.

ISO/PAS 8800 complements existing safety standards such as ISO 26262 for functional safety and ISO 21448 (SOTIF), which focus on E/E system malfunctions and inadequate system performance, respectively. By contrast, ISO/PAS 8800 introduces an AI safety lifecycle and addresses aspects such as bias, prediction accuracy, robustness, and generalization capability.

The standard presents OEMs and suppliers with technical, organizational, regulatory, and economic challenges. This is primarily because, with the AI safety lifecycle, ISO/PAS 8800 introduces an additional development model that goes beyond existing processes, but since the standard merely serves as a framework, it does not define any specific requirements. This leaves a lot of room for interpretation making the implementation in practice more difficult.

For example, existing development processes will have to be adapted without clear guidelines for implementation. Additionally, there is still much uncertainty as to how such an AI management system can be brought into compliance with existing legal regulations. Following and interpreting the standard therefore requires an investment in training, research, development, and compliance – without there being sufficiently established best practices that could serve as a guide.

Since traditional test procedures, which were established for rule-based data processing, are not sufficient for AI systems, ISO/PAS 8800 takes a comprehensive approach: Safety requirements are outlined for the entire development process. In particular, expanded measures for verification and validation, as well as a more comprehensive field monitoring after the market launch play a key role. This ensures the safety of the systems in the long term.

As complete verification of AI systems is not possible, the standard focuses on measures that cover all life cycle phases – from development and verification to field monitoring after the market launch. The aim is to ensure an adequate level of safety despite existing uncertainties.