Autonomous vehicles rely on cutting-edge technological advancements and shall provide safer, more convenient, and enjoyable commutes by introducing benefits such as reduced driver stress, diminished demand for parking at destinations, and improved accessibility for people with disabilities, to mention just a few of them. Despite the recent technological advancements, safety assurance and safety certification remain a challenge for the industry, especially when AI-based systems are part of the safety loop. Furthermore, AI-based systems could be part of CCAM, so their validation represents an important part of the validation of CCAM. This chapter deals with a fault-tolerant safety architecture, considering a runtime safety assurance approach, and investigates the challenges related to safety assurance of the sensing and perception subsystem. The AI-enabled sensing and perception subsystem, considers a multi-modal sensing approach having a 2oo3 architecture. Initially, a methodical examination is carried out to assess the resilience and responsiveness of the AI-driven sensing and perception subsystem. The primary emphasis is placed on understanding the impact of different factors on the reliability of object detection. The results demonstrate that the system’s object detection capability is influenced by factors such as object size, weather conditions, and illumination conditions, similar to the characteristics of the human eye. Subsequently, safety protocols are implemented to address potential issues such as sensor failure, non-detection of objects, or reduced sensor functionality. These protocols include fault detection, tracking multiple objects, and integrating data from multiple sensors, also known as sensor fusion. In the event of a failure in the sensing and perception subsystem, the vehicle is programmed to perform an emergency maneuver to ensure safety. This involves bringing the vehicle to a halt in a safe location, such as the emergency lane or the side of the street.