Thinking About the Future of Warfare

This research aims to evaluate the feasibility of using Large Language Models (LLMs) to create scalable and undetectable honeypots. The current challenge is scaling honeypots without making them obvious to attackers, hindered by the scarcity of specialized IT security personnel. LLMs offer a solution by potentially creating unique, hard-to-identify systems. The initial phase uses an NVIDIA RTX 3090ti and Debian to have the LLM generate insecure C++ programs with exposed ports to attract attackers. The main hurdle is expanding this to simulate a real corporate IT network by training the LLM to generate interconnected honeypots that mimic authentic business infrastructure, requiring a dataset that encapsulates realistic yet vulnerable systems and their interactions.