Generative AI-based Phishing Text Generation Using Hybrid Prompt Design with Heuristic Algorithm for Multimodal Phishing Detection
International journal of intelligent engineering and systems · 2025
In the real world, phishing attacks utilize vulnerabilities in automated systems and human behavior to pose serious security risks.Nevertheless, traditional detection systems have several shortcomings and challenges, including identifying the intricate and deceptive phishing emails that exploit generative intelligence and Uniform Resource Locator (URL) manipulation.Thus, this work addresses the gaps in traditional phishing detection systems against the evolving phishing tactics by developing a unique framework for Large Language Model (LLM)-Guided Phishing Text Generation and Detection.Initially, the proposed approach utilizes generative Artificial Intelligence (AI) to generate intelligent phishing emails using hybrid prompt engineering with prompt chaining and evaluation of a metric-aware genetic algorithm to enhance the phishing email content.Then, the URL tabular data for multimodal detection is constructed.Email data is enriched using the Genetic Algorithm (GA) to improve detection abilities, which selects highly relevant and coherent phishing descriptions through an optimization process for email enrichment.Extracting the phishing behavior-influential features from the URL structure precisely enforces multimodal phishing detection.In subsequence, by jointly learning the email content and structured URL data, the proposed approach enhances the precision of phishing detection, designed with the cross-attention associated multimodal transformer architecture.From the extensive experimental evaluations, the proposed approach yields a balanced score of recall as 96.47% and an accuracy of 96.91% by the generative phishing knowledge, accomplished by the prompt-based generation, LLM, heuristic intelligence, and URL tabular feature extraction in the phishing detection system.Finally, the results demonstrate the proposed system's effectiveness in recognizing diverse phishing texts while testing AI-enriched text.Compared to baseline models, it addresses challenges in dynamic threats and multimodal dependencies through generative AI and cross-attention transformer, respectively, ensuring robust phishing detection in diverse environments.