We explore the security vulnerabilities and reliability challenges inherent in modern software and AI-driven systems. This includes characterizing and mitigating adversarial attacks against machine learning models, particularly in reinforcement learning contexts, and analyzing potential weaknesses in data structures like graphs. Furthermore, we develop techniques such as fuzzing and automated validation to proactively identify and address potential security flaws and reliability issues in diverse systems, from multi-threaded programs to critical infrastructure components like buffer management.