Lune

EMNLP2025Top-tier venue

AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender

Weixiang Zhao, Jiahe Guo, Yulin Hu, Yang Deng, An Zhang, Xingyu Sui, Xinyang Han, Yanyan Zhao, Bing Qin, Tat-Seng Chua, Ting Liu

Abstract

Despite extensive efforts in safety alignment, large language models (LLMs) remain vulnerable to jailbreak attacks. Activation steering offers a training-free defense method but relies on fixed steering coefficients, resulting in suboptimal protection and increased false rejections of benign inputs. To address this, we propose AdaSteer, an adaptive activation steering method that dynamically adjusts model behavior based on input characteristics. We identify two key properties: Rejection Law (R-Law), which shows that stronger steering is needed for jailbreak inputs opposing the rejection direction, and Harmfulness Law (H-Law), which differentiates adversarial and benign inputs. AdaSteer steers input representations along both the Rejection Direction (RD) and Harmfulness Direction (HD), with adaptive coefficients learned via logistic regression, ensuring robust jailbreak defense while preserving benign input handling. Experiments on LLaMA-3.1, Gemma-2, and Qwen2.5 show that AdaSteer outperforms baseline methods across multiple jailbreak attacks with minimal impact on utility. Our results highlight the potential of interpretable model internals for real-time, flexible safety enforcement in LLMs.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

4 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs cites Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs cites AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models
  • MULTIGUARD: An Efficient Approach for AI Safety Moderation Across Languages and Modalities cites HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
  • MULTIGUARD: An Efficient Approach for AI Safety Moderation Across Languages and Modalities cites Jailbroken: How Does LLM Safety Training Fail?
  • MULTIGUARD: An Efficient Approach for AI Safety Moderation Across Languages and Modalities cites Refusal in Language Models Is Mediated by a Single Direction
  • AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models cites HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers4

  1. Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMsWeixiang Zhao, Yulin Hu, Yang Deng, Jiahe Guo et al.ACL 2025 · 23 citations
  2. When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue AgentsJiahe Guo, Xiangran Guo, Yulin Hu, Zimo Long et al.ACL 2026 · 5 citations
  3. Steering at the Source: Style Modulation Heads for Robust Persona ControlYoshihiro Izawa, Gouki Minegishi, Koshi Eguchi, Sosuke Hosokawa et al.ICML 2026 · 2 citations
  4. MULTIGUARD: An Efficient Approach for AI Safety Moderation Across Languages and ModalitiesSahil Verma, Keegan Hines, Jeff A. Bilmes, Charlotte Siska et al.EMNLP 2025

Builds on21

  1. Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
  2. Jailbroken: How Does LLM Safety Training Fail?Alexander Wei, Nika Haghtalab, Jacob SteinhardtNeurIPS 2023 · 2,230 citations
  3. Refusal in Language Models Is Mediated by a Single DirectionAndy Arditi, Oscar Obeso, Aaquib Syed, Daniel Paleka et al.NeurIPS 2024 · 1,166 citations
  4. HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust RefusalMantas Mazeika, Long Phan, Xuwang Yin, Andy Zou et al.ICML 2024 · 1,031 citations
  5. AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language ModelsXiaogeng Liu, Nan Xu, Muhao Chen, Chaowei XiaoICLR 2024 · 722 citations

Related papers

  1. Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against JailbreaksHan Wang, Gang Wang, Huan ZhangCVPR 2025
  2. Principled Steering via Null-space Projection for Jailbreak Defense in Vision-Language ModelsXingyu Zhu, Beier Zhu, Shuo Wang, Junfeng Fang et al.CVPR 2026 · 5 citations
  3. AlphaSteer: Learning Refusal Steering with Principled Null-Space ConstraintLeheng Sheng, Changshuo Shen, Weixiang Zhao, Junfeng Fang et al.ICLR 2026 · 52 citations
  4. Why Not Act on What You Know? Unleashing Safety Potential of LLMs via Self-Aware Guard EnhancementPeng Ding, Jun Kuang, Zongyu Wang, Xuezhi Cao et al.ACL Findings 2025
  5. Detecting What Queries Seek: Steering LLM Safety with FFN Output Activation MonitoringXiaohao Luo, Ying Wei, Rui ZhaoACL 2026
Dusk over the sea between two cliffs drawn in fine vertical lines