Jailbreaking with Universal Multi-Prompts

TL;DR

This paper introduces JUMP, a prompt-based universal attack method for jailbreaking large language models, and DUMP, a defense approach, demonstrating superior performance over existing techniques.

Contribution

The paper presents JUMP and DUMP, novel universal multi-prompt techniques for attacking and defending LLMs, addressing the generalization challenge across unseen tasks.

Findings

JUMP outperforms existing attack methods in effectiveness.

DUMP provides a robust defense against universal prompts.

Universal prompts transfer well across different tasks.

Abstract

Large language models (LLMs) have seen rapid development in recent years, revolutionizing various applications and significantly enhancing convenience and productivity. However, alongside their impressive capabilities, ethical concerns and new types of attacks, such as jailbreaking, have emerged. While most prompting techniques focus on optimizing adversarial inputs for individual cases, resulting in higher computational costs when dealing with large datasets. Less research has addressed the more general setting of training a universal attacker that can transfer to unseen tasks. In this paper, we introduce JUMP, a prompt-based method designed to jailbreak LLMs using universal multi-prompts. We also adapt our approach for defense, which we term DUMP. Experimental results demonstrate that our method for optimizing universal multi-prompts outperforms existing techniques.