Adaptive Behavioral Analysis Systems (ABAS) for Detecting URL Phishing in LLM-Based Environments

Abstract

Large Language Models (LLMs) have transformed various industries by enhancing natural language processing applications. However, their capabilities also introduce cybersecurity vulnerabilities, particularly regarding phishing attacks. Cybercriminals are using LLMs to create highly deceptive phishing URLs, making it increasingly challenging for existing detection systems to keep pace with evolving attack strategies. To address these challenges, this study introduces an Adaptive Behavioral Analysis System (ABAS) specifically designed to detect phishing URLs in LLM-based environments. ABAS combines behavioral analytics with URL feature extraction and employs preprocessing techniques such as cleaning, normalization, and tokenization to identify meaningful patterns in URLs. The model is trained and validated on a dataset of 50,000 legitimate and phishing URLs, ensuring its adaptability to real-world phishing threats. Experimental evaluations show that ABAS achieves an accuracy of 96.4%, outperforming current phishing detection systems. The results highlight ABAS’s capacity to dynamically adapt to evolving phishing tactics, providing a robust and efficient defense mechanism against LLM-generated phishing threats. This research not only uncovers vulnerabilities in LLM-based phishing attacks but also contributes to the development of adaptive and resilient cybersecurity frameworks. Future work will focus on further enhancing ABAS by integrating real-time detection and continuous learning capabilities.