Towards Transparent Phishing Email Detection: A Transformer-Based Explainable AI Approach

Phishing attacks, which frequently take advantage of user trust and ignorance, remain a leading source of cyberthreats. Even though machine learning algorithms have demonstrated promise in identifying phishing emails, consumer confidence and transparency are limited by their black-box nature. This work introduces an explainable AI (XAI) framework for phishing detection that combines counterfactual explanations with a refined BERT model. Emails are classified as either authentic or phishing by the algorithm, and the explanation component identifies the fewest adjustments needed to reverse the categorization. Our method reduced user error in threat assessment by achieving 94.2% accuracy on benchmark datasets and greatly enhancing human interpretability. The suggested solution increases phishing awareness and builds confidence in automated cybersecurity technologies by enabling users to comprehend model judgments.