AI-Based Vehicle Safety and Functional Safety Standards in Autonomous Cars: A Use Case in Safety-Critical Systems

The increasing sophistication of autonomous vehicle (AV) technologies demands the deployment of intelligent, datacentric solutions to ensure real-time safety and reliability. This study offers an in-depth review and experimental evaluation of AI-driven safety systems, with a focus on the application of deep learning (DL) and machine learning (ML) approaches in safetycritical contexts. Twenty recent and influential research works were systematically assessed, encompassing AI-powered modules for perception, control, and hazard detection in AV environments. These studies were further examined in relation to established safety frameworks, including ISO 26262, SOTIF, and UL 4600, to evaluate the regulatory readiness of AI-based strategies. Complementing the literature review, a simulation-based experiment was performed using synthetic binary classification data to assess the effectiveness of AI in identifying safety-relevant scenarios. The model demonstrated strong performance, achieving an accuracy of up to 95%, a macro-averaged F1-score of 0.73, and an AUC of 0.84. Confusion matrix analysis confirmed a high detection rate for safety events, while evaluation of inference time across DL/ML models revealed trade-offs between computational speed and interpretability, particularly in hybrid and explainable AI systems. These findings underscore the promise of DL/ML in advancing AV safety but also point to persistent issues related to explainability, response latency, and compliance with certification processes. In conclusion, this research affirms that AI technologies, when properly validated and integrated, have the capacity to significantly enhance automotive safety. Nonetheless, ongoing advancements are necessary to align AI-driven frameworks with rigorous safety engineering and certification standards. This paper contributes actionable insights that bridge theoretical advancements with practical AV safety applications.