Smart cities are rapidly embracing interconnected IoT devices and critical infrastructure, expanding the complexity and vulnerability of urban cyberspace. Conventional cybersecurity remains largely reactive and signature-based, falling short against advanced attacks such as zero-day exploits and polymorphic malware. This research introduces a novel AI-driven threat intelligence framework for predictive cyber defense in smart city environments that integrates federated learning for privacy-preserving analytics, blockchain-secured data provenance for immutable logging, and real-time anomaly detection to anticipate and mitigate threats across municipal networks. The fusion of behavioral baselining and global threat feeds enables accurate identification of sophisticated attacks while ensuring responsible AI governance through transparent, explainable protocols. Empirical validation with real-world smart city deployments demonstrates 99.47% threat detection accuracy and a 40% reduction in incident response time. Distinctly, the framework showcases optimization and industrial engineering impacts by refining resource allocation, minimizing workload for SOC analysts, and enhancing resilience of citywide operational systems. This explicit linkage to data science and industrial engineering underscores the study’s relevance to conference themes, highlighting contributions to resource optimization and automation of urban critical infrastructures. Overall, the proposed solution supports robust urban cybersecurity aligned with user trust and regulatory mandates, setting a new benchmark for sustainable and resilient smart cities.