arXiv cs.CL· Xiaoqun Liu, Weiming Qi, Qiben Yan·· 13 小时前AI 评分34
DYNASHIELD：通过动态解码定制为 LLM 提供黑盒移动目标防御
DYNASHIELD: A Black-Box Moving Target Defense for LLMs via Dynamic Decoding Customization
AI 导读
DYNASHIELD 是一个面向黑盒 LLM 部署的移动目标防御框架，通过在推理时定制解码超参数和系统提示词提升鲁棒性，无需重训练或访问模型内部。该方法先筛选能降低攻击成功率的解码配置，再从加权配置池中概率采样以引入受控的行为变化。在 7 个开源 LLM 和 4 种 SOTA 越狱攻击（AdvBench 对抗提示词）上，其攻击成功率较 7 种基线防御大幅下降，同时保持回复质量且推理开销极小。
正文 · 原文
Abstract:Large language models (LLMs) remain vulnerable to jailbreak attacks in which adversarial prompts induce harmful outputs. Existing defenses often require access to the model internals or additional training, limiting their applicability for service providers deployed through black-box APIs. In this paper, we propose DYNASHIELD, a moving target defense framework that improves robustness by customizing decoding hyperparameters and system prompts at inference time. DYNASHIELD includes two key steps: (1) it identifies decoding configurations that reduce attack success probability, and (2) it probabilistically samples from a weighted configuration pool to introduce controlled variability in model behavior. We evaluate DYNASHIELD across 7 open-source LLMs under 4 state-of-the-art jailbreak attacks, using adversarial prompts from AdvBench. Results show substantial reductions in attack success rate compared with 7 baseline defenses, while maintaining response quality and incurring minimal inference overhead. Because DYNASHIELD operates solely through exposed runtime controls, it requires no retraining or model-internal access. These results suggest that safety-aware dynamic decoding is a promising and practically lightweight defense mechanism for black-box LLM deployments.
| Comments: | Accepted by The 29th International Symposium on Research in Attacks, Intrusions and Defenses (RAID) 2026 |
| Subjects: | Cryptography and Security (cs.CR); Computation and Language (cs.CL) |
| Cite as: | arXiv:2412.07672 [cs.CR] |
| (or arXiv:2412.07672v3 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2412.07672 arXiv-issued DOI via DataCite |
Submission history
From: Qiben Yan [view email]
[v1]
Tue, 10 Dec 2024 17:02:28 UTC (1,183 KB)
[v2]
Sun, 16 Aug 2026 15:06:14 UTC (1,496 KB)
[v3]
Tue, 27 Oct 2026 14:52:52 UTC (1,496 KB)
来源：arXiv cs.CL · arxiv.org