Meta GPT-Based Agent for Enhanced Phishing Email Detection

Phishing emails are a common means of phishing and their detection is an important aspect of email security. In this study, Meta Phishing Detector Agent system is designed, which is constructed based on Meta GPT framework and uses LLMs as the core of agent to detect phishing emails. The agent has three actions: Check Header, Check Body and Make Decision. Check Header and Check Body are used to analyze the header and body information of emails, and Make Decision is responsible for giving structured decision results based on the previous analysis results. Unlike previous email classification systems, our system provides users with detailed evidence-based reasons for judgment while giving judgment results. The performance of the system was evaluated on a novel phishing email dataset, and the experimental results show that the agent system, using the GLM-4 large language model for example, achieves 99.30% accuracy. Our study gives a new practical approach for the application of LLMs-based agents in the field of cyber security.