From online advertising to content delivery networks and recommendation systems, many modern technologies rely on algorithms that must operate in real time without knowing what will happen next. Researchers in computer science, operations research, engineering, and other fields have developed powerful online optimization and learning tools for making effective decisions in the face of uncertainty, by helping systems learn from past outcomes to improve performance over time. However, these methods are vulnerable to attackers aiming to disrupt the system: fake reviews, ad fraud, denial-of-service attacks, and other attacks can corrupt the algorithms' learning process. Researchers have begun developing "corruption-robust" learning algorithms that are more resilient to attacks; however, significant barriers remain in translating these theoretical advances into real-world systems. This project aims to reduce those barriers by designing learning algorithms that are both theoretically sound and practical to implement, enabling more robust decision-making in real-world applications. This work directly supports the national interest by strengthening the resilience of critical cyberinfrastructure and advancing the scientific foundations of trustworthy AI. This project advances the theoretical foundations of online optimization and learning by explicitly incorporating robustness to security threats and adversarial corruptions. While online learning has been extensively studied for