Leveraging Large Language Models to Obscure Code Stylometry: A Comparative Study of GPT-3.5 and GPT-4

This study examines the use of Large Language Models (LLMs), specifically GPT-3.5 and GPT-4, to obscure code stylometry for authorship attribution and cybersecurity. It evaluates the models' ability to modify code while preserving functionality, employing various prompt engineering strategies, and demonstrates significant differences in effectiveness between single-shot and multi-shot prompting. The findings underscore the challenges in maintaining code integrity and the implications for authorship detection techniques in the context of advanced AI, which is crucial for practitioners in cybersecurity and software engineering.