Autonomous vehicles (AVs) powered by artificial intelligence promise a fundamental transformation of road transportation, with potential benefits ranging from near-elimination of human-error-induced crashes to enhanced mobility for populations unable to drive. Yet despite substantial engineering progress and billions of kilometres of cumulative testing, the question of whether computerized self-driving cars are genuinely safe remains technically nuanced and far from resolved. This review critically examines the current state of AI-based autonomous vehicle safety by synthesizing findings from three recent peer-reviewed studies: (1) Grewal et al. [2024/2025], who evaluate Bayesian uncertainty quantification (UQ) methods for predicting safety-critical misbehaviours in simulation-based ADS testing; (2) Nouri et al. [2024], who propose an LLM-based pipeline for automating Hazard Analysis and Risk Assessment (HARA) under ISO 26262 and SOTIF; and (3) Ullrich et al. [2024], who survey the intersection of AI safety research, standardization, and regulation for automated vehicles.