Autonomous driving technologies are rapidly advancing through AI-based hardware and software, yet their decision-making processes are still widely perceived as black boxes. This presentation explores how AI-based autonomous driving systems can be interpreted and validated from the perspective of Explainable AI (XAI). It raises the key question of whether explainability and testability can be achieved across the perception, decision-making and control layers even within end-to-end learning architectures, and presents practical validation strategies using simulation- and data-driven testing. Through real-world collaboration cases, the presentation demonstrates how trust, safety and regulatory readiness of autonomous driving AI can be strengthened in practice.