Explainable artificial intelligence for cyber-resilient smart cities: enhancing trust, transparency, and adversarial robustness

Abstract

The rapid integration of Artificial Intelligence (AI) into smart city infrastructures has transformed urban services, enabling intelligent transportation, energy optimization, predictive governance, and advanced cybersecurity mechanisms. However, the increasing complexity of AI-driven systems has also introduced critical vulnerabilities related to adversarial attacks, opaque decision-making, and reduced public trust. Within the context of the AIAS project, which focuses on AI-assisted cybersecurity and adversarial AI resilience, this paper investigates the role of Explainable Artificial Intelligence (XAI) in strengthening cyber-resilient intelligent cities. The research explores how explainability mechanisms can improve the transparency, accountability, and interpretability of AI-based cybersecurity systems operating in smart urban environments. The study examines state-of-the-art XAI techniques, including feature attribution methods, surrogate models, attention visualization, and human-centered interpretability frameworks, applied to cyber threat detection and anomaly monitoring systems. Particular attention is given to adversarial AI scenarios, where malicious manipulation of machine learning models may compromise decision integrity and urban critical infrastructure. The paper proposes an explainability-oriented framework capable of supporting both “AI for Cybersecurity” and “Cybersecurity for AI” approaches by enabling security analysts and public administration stakeholders to better understand automated decisions, identify anomalies, and validate threat intelligence outputs. The framework aims to reduce the “black-box” nature of deep learning systems while supporting compliance with emerging regulatory requirements regarding trustworthy and ethical AI governance. Expected outcomes include improved trust in AI-enabled public services, enhanced incident response capabilities, and increased resilience of smart city ecosystems against adversarial threats and information sabotage. Furthermore, the research contributes to the broader discussion on ethical AI deployment in public administration by emphasizing transparency, human oversight, and responsible data governance. The study aligns with current European priorities on trustworthy AI and cyber resilience, supporting the development of secure, explainable, and citizen-centered intelligent urban infrastructures.