Efficient LLM Jailbreak via Adaptive Dense-to-sparse Constrained Optimization

TL;DR

This paper presents ADC, a novel continuous optimization-based method for effectively jailbreaking open-source LLMs by gradually increasing sparsity, outperforming existing token-level attack techniques in success rate and efficiency.

Contribution

Introduces ADC, a new token-level attack method that relaxes discrete optimization into a continuous process with increasing sparsity, improving jailbreak success rates.

Findings

Achieves highest attack success rate on 7 out of 8 LLMs

More effective and efficient than state-of-the-art methods

Bridges discrete and continuous optimization in token attacks

Abstract

Recent research indicates that large language models (LLMs) are susceptible to jailbreaking attacks that can generate harmful content. This paper introduces a novel token-level attack method, Adaptive Dense-to-Sparse Constrained Optimization (ADC), which has been shown to successfully jailbreak multiple open-source LLMs. Drawing inspiration from the difficulties of discrete token optimization, our method relaxes the discrete jailbreak optimization into a continuous optimization process while gradually increasing the sparsity of the optimizing vectors. This technique effectively bridges the gap between discrete and continuous space optimization. Experimental results demonstrate that our method is more effective and efficient than state-of-the-art token-level methods. On Harmbench, our approach achieves the highest attack success rate on seven out of eight LLMs compared to the latest jailbreak methods. Trigger Warning: This paper contains model behavior that can be offensive in nature.