Diversity Helps Jailbreak Large Language Models

TL;DR

This paper introduces a novel jailbreak technique exploiting LLMs' ability to diverge from context, significantly increasing success rates in bypassing safety measures across multiple chatbots with fewer queries.

Contribution

The authors reveal a new method that outperforms existing jailbreak approaches by leveraging divergence and obfuscation, exposing vulnerabilities in current LLM safety training.

Findings

Achieves up to 62.83% higher success rate in bypassing chatbots

Uses only 12.9% of the queries compared to previous methods

Exposes critical flaws in current LLM safety measures

Abstract

We have uncovered a powerful jailbreak technique that leverages large language models' ability to diverge from prior context, enabling them to bypass safety constraints and generate harmful outputs. By simply instructing the LLM to deviate and obfuscate previous attacks, our method dramatically outperforms existing approaches, achieving up to a 62.83% higher success rate in compromising ten leading chatbots, including GPT-4, Gemini, and Llama, while using only 12.9% of the queries. This revelation exposes a critical flaw in current LLM safety training, suggesting that existing methods may merely mask vulnerabilities rather than eliminate them. Our findings sound an urgent alarm for the need to revolutionize testing methodologies to ensure robust and reliable LLM security.