In this paper, we introduce the first-ever graph neural network (GNN) and temporal logic (TL)-based real-time anomaly detection method for IoT-enabled cyber-physical systems. The GNN-TL model offers an effective framework for detecting and analysing anomalies in complex, interconnected environments. The results, evaluated using various metrics such as detection accuracy, false positives and false negatives, demonstrate the model’s strong overall performance compared to traditional approaches. Moreover, by incorporating temporal logic into model checking, the detection of time-dependent anomalies is enhanced, underscoring its importance in understanding system behaviour. These findings reaffirm the capability and value of the GNN-TL model for practical use, providing robust and timely anomaly detection that can improve the security and resilience of IoT-enabled systems in real-time.