Generative AI and deep learning for phishing detection: a comparative analysis

Phishing attacks remain a major cybersecurity threat, necessitating advanced detection frameworks. This paper compares phishing detection techniques using LSTM networks, multi-attention models, and generative AI models like BERT and RoBERTa. We evaluate their performance in detecting sophisticated phishing attempts, focusing on adaptability to evolving tactics and adversarial resilience. Additionally, we explore emerging concepts such as multi-modal detection and hybrid approaches for future advancements. Our research highlights that combining deep learning’s sequence awareness with generative AI’s contextual understanding improves detection accuracy and adaptability. Hybrid methodologies, integrating discriminative and generative models, prove more effective in enhancing phishing detection and developing robust security solutions.