Romanian Cyber Security Journal / Spring 2024, No. 1, Vol. 6

Shared Prime Vulnerability Analysis for Pseudo- Random Number Generator Implementation Proposed by AI Models

Ajay Singh RATHORE, Samuel TĂNASE

---

Abstract

Large Language Models have become popular for performing different tasks like summarization, translations, classification and code generation. This has led to different types of research where models are observed and their architecture is improved for better performance, including the tasks mentioned above. Other studies branch into finding a better way to generate responses through these LLMs while treating them as a black box. There are studies where the models are specially analyzed for the code generation task. We follow this category of studies to analyze the generated code for security weaknesses like CWEs. A few different publicly available models were used and prompted to provide the implementation of pseudo-random number generation for prime numbers. The purpose of this study is to find out if these generated prime numbers will be secure enough to be used for cryptography algorithms such as RSA.

Keywords

Security Weaknesses, Large Language Models, Code Generation, Security Weaknesses,, CWEs, RSA