AI-Driven Malware Detection: Leveraging LLM - Based Behavioral Embeddings for Cybersecurity

Malware remains a serious threat to critical infrastructure, resulting in data breaches and financial damage in sectors like healthcare, energy, and transportation. As malware techniques become more advanced, traditional detection methods like signature-based and heuristic approaches often fail to keep up, especially when confronting zero-day threats. While machine learning and deep learning have shown potential, they still rely heavily on manual feature engineering for input data, which often limits their ability to generalize to unseen or evolving attacks. Recent advancements in large language models (LLMs) offer promising new direction for improving malware detection by embedding and analyzing complex static and dynamic features. This study introduces a novel AI-driven framework that leverages behavioral embeddings generated by large language models (LLMs) for multi-class malware classification. The model is trained and evaluated on the Avast-CTU Public CAPEv2 dataset. Experimental results show strong performance, highlighting the effectiveness of LLM-based embeddings in capturing behavioral patterns for accurate malware classification.