Sarcouncil Journal of Multidisciplinary

Abstract: The fast adoption of machine learning (ML) into cyber-physical systems (CPS) is changing the operations of most critical infrastructure in the energy, utilities, and safety-critical industries. On one hand, AI-based predictive analytics and adaptive control can be used to improve efficiency and detect faults; however, it also introduces new weaknesses concerning the integrity of the data, the robustness of the models, adversarial manipulation, and the policy controls. The layered risks can be tackled only with the help of traditional cybersecurity methods. This review contributes to an ML-integrated CPS resilience-by-design framework based on the principles of resilience engineering and in line with the United States (U.S.) cybersecurity and AI governance frameworks. It combines ideas of adaptive systems, adversarial threat models, organizational preparedness variables, and policy frameworks to suggest a layered architectural model incorporating security, robustness, human supervision, and governance integration. The research establishes critical gaps in research and future perspectives on how to operationalize resilience metrics, adversarial stress testing, and maturity-based governance alignment. Resilience-by-design offers a disciplined journey towards achieving machine learning-affiliated critical infrastructure in a shifting threat environment by introducing engineering, policy, and socio-technical viewpoints into the solution.