Conceptual

Generative AI and Structured Prompt Engineering for Adaptive Cyber Deception

SPADE, a systematic framework that uses generative-AI language models driven by structured prompt engineering to automatically generate adaptive cyber-deception ploys (such as honeypot and decoy configurations) tuned to observed malware behavior. It targets the specific failure modes large language models bring to this task, generalized outputs, ambiguity, under-use of contextual information, and limited scalability, and evaluates candidate models across malware scenarios using Recall, Exact Match, BLEU, and expert quality assessments, finding ChatGPT-4o the strongest performer.