Ensuring AI Safety in Autonomous Vehicles: A Framework Based on ISO PAS 8800

This study presents a structured exploration of ISO PAS 8800 as a dedicated safety framework addressing the unique challenges posed by artificial intelligence (AI) in autonomous vehicles (AVs). The research aims to establish the necessity of a distinct safety standard beyond conventional protocols, such as ISO 26262 and ISO 21448, which are insufficient to manage the probabilistic, adaptive, and opaque characteristics inherent in AI- driven systems. Employing a qualitative methodological approach grounded in standards analysis and case-based synthesis, the study evaluates the provisions of ISO PAS 8800 across multiple dimensions, risk governance, system transparency, continuous validation, and human oversight. Key findings demonstrate that ISO PAS 8800 fills critical gaps left by existing safety standards, offering AI-specific safety lifecycle processes, interpretability protocols, and robust risk management strategies. It intro- duces novel concepts such as Component Fault and Deficiency Trees (CFDTs), scenario-based validation, bounded incremental learning, and post-deployment monitoring, which are essential for certifying learning-enabled and continuously evolving AV systems. Furthermore, the framework emphasizes harmonization with cybersecurity standards (e.g., ISO/SAE 21434) to address adversarial vulnerabilities in AI pipelines. ISO PAS 8800 provides a comprehensive, adaptable, and forward-compatible framework for the governance of AI safety in autonomous driving. It facilitatesthe development of trustworthy, auditable, and socially accountable AV technologies, aligning technical innovation with emerging regulatory and ethical expectations.