Artificial Intelligence is transforming cyber-physical systems by enabling automation, predictive insights, and adaptive control across critical domains such as utilities, transportation, and manufacturing. Yet the increasing complexity of these systems, and their exposure to adversarial and environmental uncertainties, demands AI approaches that are not only accurate but also trustworthy, resilient, and context-aware. In this talk, I introduce a context-driven deep learning framework that integrates diverse sources of information with advanced neural architectures to support reliable short-term forecasting in dynamic environments. By incorporating both internal system signals and external contextual factors, this approach enhances decision support, strengthens resilience during critical conditions, and reduces reliance on purely reactive operations. I will highlight the design principles behind this work, discuss its application to real-world cyber-physical infrastructures, and outline how such methods can inform the next generation of trustworthy AI pipelines. Looking ahead, my research agenda at 98ÌÃ’s School of Cybersecurity focuses on advancing AI safety, security, and privacy through context-aware, adversarially robust, and explainable learning systems. This vision aims to bridge theory and practice, ensuring that AI-driven cyber-physical systems remain secure, ethical, and reliable in the face of evolving challenges.