Abstract
This paper describes the evolution of malware presents improvement of challenges into cybersecurity with traditional signature-based detection often failing to predict and identify the zero -day and polymorphic attacks in the system. we proposed novel attention-based framework for malware detection that leverages the structural and behavioral character tics of the executable files in the system. the attention mechanism with GAT models enables selective focus on critical nodes and connections capturing both local execution patterns and global behaviors structure which are of are often indicative of malicious activity. To improve the transparancy and trust in automated malware protection system, the proposed system framework connect with explainable AI system including SHAP analysis and attention weight visualization of the system it allows cybersecurity analyst to interpret which API sequences or interactions contributed the most significant to models decision, facilizing actionable insights for threat control and mitigation. here we done the evaluation on a benchmark malware dataset demonstrates that the GAT based model achieves high detection accurate, precision and recall, while maintaining low false positive rates. The combination of structure graph modeling, attention based learning and explain ability provides a robust and interpretable malware detection and solutions suitable for deployment in dynamic cyber security environments.