Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM

TL;DR

This paper proposes a method to enhance the robustness of aligned large language models against adversarial and jailbreaking prompts without retraining, significantly reducing attack success rates.

Contribution

Introduction of RA-LLM, a robust alignment checking framework that defends against alignment-breaking attacks without retraining the original LLM.

Findings

RA-LLM reduces attack success rates from nearly 100% to around 10% or less.

Theoretical analysis confirms RA-LLM's effectiveness in defending against attacks.

Experimental results on open-source LLMs demonstrate improved robustness.

Abstract

Recently, Large Language Models (LLMs) have made significant advancements and are now widely used across various domains. Unfortunately, there has been a rising concern that LLMs can be misused to generate harmful or malicious content. Though a line of research has focused on aligning LLMs with human values and preventing them from producing inappropriate content, such alignments are usually vulnerable and can be bypassed by alignment-breaking attacks via adversarially optimized or handcrafted jailbreaking prompts. In this work, we introduce a Robustly Aligned LLM (RA-LLM) to defend against potential alignment-breaking attacks. RA-LLM can be directly constructed upon an existing aligned LLM with a robust alignment checking function, without requiring any expensive retraining or fine-tuning process of the original LLM. Furthermore, we also provide a theoretical analysis for RA-LLM to verify its effectiveness in defending against alignment-breaking attacks. Through real-world experiments on open-source large language models, we demonstrate that RA-LLM can successfully defend against both state-of-the-art adversarial prompts and popular handcrafted jailbreaking prompts by reducing their attack success rates from nearly 100% to around 10% or less.