ChatSpamDetector: Large Language Model-Based Phishing Email Detection

Phishing emails continue to pose a serious cybersecurity threat, even though modern email services employ spam filtering, sender authentication, and content-based defenses.In practice, phishing messages may still evade detection, while legitimate emails may be incorrectly flagged as suspicious.Such limitations highlight the need for detection methods that are not only accurate but also explainable to end users.In this paper, we present ChatSpamDetector, a phishing email detection system based on Large Language Models (LLMs).The system analyzes email headers and body text and produces both a phishing classification and a human-readable rationale for the decision.We evaluate ChatSpamDetector on datasets of phishing and legitimate emails and compare its performance with multiple LLMs and baseline approaches.Experimental results show that the proposed system achieves 99.70% accuracy and that prompt design substantially affects detection performance.We also analyze the generated rationales to examine how the models use contextual evidence in email content and metadata.The results suggest that LLM-based phishing detection can complement existing email security mechanisms by combining strong detection capability with interpretable explanations.This combination can help users better understand warning decisions and make more informed judgments when handling suspicious emails.