Utilizing Large Language Models with Human Feedback Integration for Generating Dedicated Warning for Phishing Emails

With the rise of digital communication, phishing has emerged as the predominant cybercrime. Automated detection systems encounter challenges such as user trust issues and false positives, while human-centric solutions are resource-intensive and struggle with sophisticated attacks. Despite this threat, research on empowering users with automatic anti-phishing systems remains limited. This paper introduces a human-centric framework that utilizing Large Language Models (LLMs) to extract phishing indicators and generate meaningful warnings. Recognizing that certain information is unique to users, our system integrates user insights into anti-phishing measures. Preliminary results demonstrate the promise of LLM-driven approaches in crafting meaningful warnings, highlighting the synergy between human insight and machine intelligence in combating phishing. Our framework achieves over 80% effectiveness in identifying phishing semantics with no false positives or negatives, indicating high precision. This research represents a significant advancement in phishing defense, offering a nuanced and effective email security approach.