An Explainable Transformer-based Model for Phishing Email Detection: A Large Language Model Approach

AI can help protect people from phishing attacks: The paper demonstrates that a fine-tuned transformer-based model (DistilBERT) can effectively detect phishing emails with high accuracy, offering a practical tool to help organizations and individuals identify deceptive emails before they cause harm.

Explainability is essential for trust in security AI: The researchers used Explainable-AI techniques (LIME and Transformer Interpret) to show how the model makes its phishing detection decisions. This transparency is crucial for security applications where users and organizations need to understand why an email was flagged as suspicious rather than simply trusting a "black box" verdict.

Data quality and balance matter for responsible AI: The paper emphasizes that preprocessing techniques were used to clean the phishing email dataset and address class imbalance issues. This highlights that responsible AI systems require careful attention to data preparation—biased or imbalanced training data can lead to unreliable or unfair predictions.

Advanced language models can be adapted for specific safety problems: Rather than building detection systems from scratch, the researchers fine-tuned an existing large language model (DistilBERT) for the specific task of phishing detection, demonstrating how general-purpose AI tools can be responsibly repurposed to address real-world security threats.