Prompt engineering for detecting phishing
Darrell L. Young, Eric C. Larson, Mitchell Aaron Thornton · 2025
This paper introduces an adversarial framework using two Large Language Models (LLMs) tailored through prompt engineering to advance phishing detection capabilities. The first LLM operates as a generator, crafted prompts guiding it to produce sophisticated phishing emails that mimic legitimate interactions. The second LLM acts as a discriminator, with prompts designed to enhance its ability to detect and classify these generated phishing attempts accurately from genuine emails. This setup leverages the dynamic capabilities of prompt engineering to refine the models’ responses, facilitating an ongoing evolution in both phishing simulation and detection. By implementing this methodology, we aim to improve the robustness and adaptability of AI-driven security defenses against complex cyber threats.