In today's digital world, cybersecurity has become more important than ever, as threats that includes ransomware, phishing attacks and advanced persistent threats (APTs), continue to grow in complexity. Traditional malware detection methods are struggling to keep up, making it clear that more advanced, intelligent solutions are needed. To address this, an AI-driven approach to threat detection has been developed with a focus on both accuracy and interpretability. This work leverages Graph Convolutional Networks (GCNs) to identify and predict malicious behavior. A web-based platform was built to allow users to upload files, with Locality-Sensitive Hashing (LSH) used to create digital fingerprints of those files. These are then analyzed by trained GCN models to detect potential threats. Explainable AI (XAI) techniques are incorporated to ensure that the model's decision-making process remains transparent, interpretable, and accessible to both developers and end-users. To ensure the system's decisions are understandable, LIME (Lo- cal Interpretable Model-agnostic Explanations) is applied to highlight the important features that influence each prediction. Large Language Models (LLMs) are integrated to analyze threat contexts and generate insights that support better threat mitigation. Graph Attention Networks (GATs) are utilized to identify anomalous nodes within the cybersecurity graph by assigning adaptive attention weights to neighboring nodes, thereby enabling more precise detection of irregular or suspicious patterns in the network structure. This combined approach not only improves proactive cybersecurity measures but also makes the system's inner workings more transparent and trustworthy. © 2025 IEEE.