Improved Techniques for Optimization-Based Jailbreaking on Large Language Models
About
Large language models (LLMs) are being rapidly developed, and a key component of their widespread deployment is their safety-related alignment. Many red-teaming efforts aim to jailbreak LLMs, where among these efforts, the Greedy Coordinate Gradient (GCG) attack's success has led to a growing interest in the study of optimization-based jailbreaking techniques. Although GCG is a significant milestone, its attacking efficiency remains unsatisfactory. In this paper, we present several improved (empirical) techniques for optimization-based jailbreaks like GCG. We first observe that the single target template of "Sure" largely limits the attacking performance of GCG; given this, we propose to apply diverse target templates containing harmful self-suggestion and/or guidance to mislead LLMs. Besides, from the optimization aspects, we propose an automatic multi-coordinate updating strategy in GCG (i.e., adaptively deciding how many tokens to replace in each step) to accelerate convergence, as well as tricks like easy-to-hard initialisation. Then, we combine these improved technologies to develop an efficient jailbreak method, dubbed I-GCG. In our experiments, we evaluate on a series of benchmarks (such as NeurIPS 2023 Red Teaming Track). The results demonstrate that our improved techniques can help GCG outperform state-of-the-art jailbreaking attacks and achieve nearly 100% attack success rate. The code is released at https://github.com/jiaxiaojunQAQ/I-GCG.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Token-forcing loss optimization | Random targets Held-out (val) | Qwen-2.5-7B Loss5.34 | 56 | |
| Jailbreak Attack | AdvBench 150 Harmful Behaviors | ASR100 | 45 | |
| Adversarial Attack against SeeAct agent | Mind2Web 600 tasks (test) | ASR Finance (pass@10)3.5 | 24 | |
| Adversarial Attack against WebExperT agent | Mind2Web 600 tasks (test) | ASR (Finance, pass@10)2.9 | 24 | |
| Jailbreak Attack | Qwen2.5-7B | Normalized Rate (NR)0.02 | 20 | |
| Jailbreak Attack | Mistral-7B | NR40 | 20 | |
| Jailbreak Attack | DeepSeek | NR Score0.00e+0 | 20 | |
| Jailbreaking Attack | MM-SafetyBench | Attack Success Rate (ASR)90 | 20 | |
| Jailbreak Attack | Gemma 4B 3 | NR32 | 20 | |
| Jailbreak Attack | GLM-4-Air | NR6 | 20 |