Demo: LLM-Based Social Engineering Detection System (LSED)

Current statistics indicate that social engineering is among the most commonly used strategies in cyber-attacks. The Federal Bureau of Investigation’s 2023 Internet Crime Report states that a total of 3.79 million complaints were filed, resulting in an estimated loss of $37.4 billion [1]. With the ongoing advancement of large language models (LLMs), one suggested approach to counter social engineering attempts is to utilize LLMs as a defense mechanism. The proposed solution leverages LLMs to detect and isolate sophisticated social engineering attempts by employing a context-based approach. This can be accomplished by processing incoming emails through a prompted LLM, which then assigns a risk score to each message based on its content. To enhance detection accuracy, this approach will incorporate Retrieval-Augmented Generation (RAG), leveraging external phishing email samples to provide additional context for the LLM.