Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM

About

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.

Bochuan Cao, Yuanpu Cao, Lu Lin, Jinghui Chen• 2023

Related benchmarks

TaskDatasetResultRank
Instruction FollowingAlpacaEval
Win Rate46.8
420
Jailbreak DefenseAdvBench
ASR (PAIR)1.7
115
Jailbreak DefenseHarmBench
PAIR ASR2.51
91
Jailbreak AttackAdvBench 150 Harmful Behaviors
ASR0.00e+0
45
Jailbreak Defense PerformanceJailbreak Attack Dataset
DSR78.6
33
Defense against adaptive attacksHarmBench
ASR5.6
28
Adversarial Attack DefenseGCG Individual
BAR99.3
18
Question AnsweringMS MARCO randomly selected 150 data points
BAR99
14
Jailbreak Defense EfficiencyHarmBench
Additional Tokens1
8
Jailbreak DefenseAdvBench Qwen-7B-chat
Attack Success Rate (ASR)15.8
7
Showing 10 of 16 rows

Other info

Follow for update