This study presents a novel approach to phishing email detection, leveraging artificial neural networks (ANN) with soft attention in natural language processing (NLP) through the integration of BERT encoders. Addressing the critical need for effective aritifical intelligence (AI)-driven cybersecurity solutions, this research combines BERT’s NLP capabilities with a modified crayfish optimization algorithm (COA) to fine-tune the hyperparameters of deep neural network models, enhancing classification accuracy. Experimental results show that our optimized model achieves a phishing detection accuracy of 92.5%, outperforming several high-performing optimizers. Comparative analysis demonstrates that this approach offers superior detection capabilities, underscoring its potential for real-world applications. This work advances the field by refining BERT’s application with optimization algorithms and provides a valuable framework for future cybersecurity research.