Evaluation empirique de la s\'ecurisation et de l'alignement de ChatGPT et Gemini: analyse comparative des vuln\'erabilit\'es par exp\'erimentations de jailbreaks

TL;DR

This paper compares the security and alignment of ChatGPT and Gemini LLMs, analyzing their vulnerabilities to jailbreak attacks through experimental evaluations and proposing a taxonomy of jailbreak techniques.

Contribution

It provides a novel comparative analysis of two major LLMs' vulnerabilities and introduces a taxonomy of jailbreak methods based on experimental results.

Findings

Gemini shows different vulnerability patterns compared to ChatGPT.

Jailbreak techniques vary in effectiveness across models.

The study highlights specific security weaknesses in both models.

Abstract

Large Language models (LLMs) are transforming digital usage, particularly in text generation, image creation, information retrieval and code development. ChatGPT, launched by OpenAI in November 2022, quickly became a reference, prompting the emergence of competitors such as Google's Gemini. However, these technological advances raise new cybersecurity challenges, including prompt injection attacks, the circumvention of regulatory measures (jailbreaking), the spread of misinformation (hallucinations) and risks associated with deep fakes. This paper presents a comparative analysis of the security and alignment levels of ChatGPT and Gemini, as well as a taxonomy of jailbreak techniques associated with experiments.