AI becomes increasingly important in safetycritical systems, especially in the case of autonomous systems, since navigation relies on AI for object detection and collision avoidance. However, safety-critical systems must adhere to functional safety standards that enforce software to be correct-by-construction, component decomposition to simplify design and validation, and the use of data only for testing purposes not to design the system itself. AI in general, and Deep Learning (DL) in particular have opposed characteristics since they have error rates (e.g., due to mispredictions), AI/DL modules can only be designed and validated monolithically, and they build on data for their design (i.e. for training purposes). Hence, DL solutions are at odds with the development process of safetycritical systems. A number of standards have recently emerged in different domains to reconcile the requirements of safety-critical systems with the characteristics of DL solutions, such as ISO 21448, ISO/IEC TR 5469, and ISO 8800, among others. However, there is a lack of realistic practice to design a DL-based safety-critical system in accordance with those regulations, and existing solutions only cover some aspects in isolation, and are often incompatible among them. SAFEXPLAIN is a 3-year Horizon Europe project addressing this challenge. SAFEXPLAIN, which finishes in September 2025, has already reached its main goals providing specific and complementary solutions to all those challenges so that AIbased safety-critical systems can be designed, implemented and validated adhering to the relevant functional safety standards in domains such as automotive, space and railway. In particular, SAFEXPLAIN provides the concepts, processes, tools and frameworks addressing the challenge end-to-end, from concept to solution. This is proven by the successful application of the SAFEXPLAIN approach in three case studies from the automotive, space and railway domains, whose results will see the light very soon.

Artificial Intelligence (AI) has traditionally prioritized high performance over compliance with functional safety standards such as IEC 61508. However, when AI systems are used in safety-related functions, it is essential to demonstrate that errors will not lead to malfunctions. This involves preventing systematic design-time errors and detecting and controlling runtime faults, as specified in IEC 61508. Moreover, ISO/PAS 8800 requires analyzing AI-specific development tools to identify and mitigate potential risks. In this paper, we take a step toward a safe end-to-end AI framework by focusing on systematic error avoidance in the implementation of You Only Look Once (YOLO), a widely used object detection model. A C-based version of YOLO-built on the Darknet framework-is analyzed using the Polyspace static analysis tool to assess MISRA C compliance. We apply corrective actions to eliminate violations, producing a MISRA $\mathbf{C}$-compliant implementation. In addition, we propose a runtime error detection mechanism using dual execution on a diverse platform and validate behavioral consistency using the COCO dataset. This approach supports the development of trustworthy AI systems by addressing both systematic errors and runtime detection.

With the advent of next-generation safety-related systems, different industries face multiple challenges in ensuring the safe operation of these systems according to traditional safety and assurance techniques. The increasing complexity that characterizes these systems hampers the maximum achievable test coverage during system verification and, consequently, it often results in untested behaviors that hinder safety assurance and represent potential risk sources during system operation. In the context of paving the way towards quantifying the risks caused by software malfunction and, hence, towards the safety-compliance of next-generation safety-related systems, this paper studies and provides a method to estimate the probability of Linux kernel execution paths that remain unobserved during the test campaign.