Safe AI Compliance in CI/CD for Software-Defined Vehicles - Researching Conformity to ISO/PAS 8800:2024

Abstract

As software-defined vehicles (SDVs) are incorporating artificial intelligence (AI), ensuring AI safety presents significant challenges for development teams. This thesis addresses four research questions spanning the current application of ISO/PAS 8800:2024 Road vehicles — Safety and artificial intelligence in SDV development, the compliance challenges teams face, where in the Continuous Integration/Continuous Delivery (CI/CD) challenges arise, and how to mitigate them. A two-phase qualitative interview study was conducted with industry professionals, including do main experts and researchers. Through semi-structured interviews, recommendable actions were produced and evaluated in the form of artefacts and a set of guidelines. The results reveal that while some individual practitioners are aware of ISO/PAS 8800:2024, formalised organisational conformance frameworks are largely absent, and a variety of different conformance issues surface throughout the entire CI/CD process rather than at any single stage. Artefacts were developed for supporting conformance to ISO/PAS 8800:2024 within existing development practices. The first artefact guides teams on where human oversight is required, where AI could be helpful, and where AI automation is feasible across the AI safety lifecycle, including the trade-offs and acceptance of AI automation at each stage. The second artefact shows which selected clauses of the ISO/PAS 8800:2024 are most prominent at each phase of the MLOps lifecycle. Integrating ISO/PAS 8800:2024 into CI/CD requires conformance to be treated as an inherent part of development rather than an after-the-fact activity. Insights from this study could be valuable for organisations integrating ISO/PAS 8800:2024 into their development workflows.