Cover Image

Steering Externalities: Benign Activation Steering Unintentionally Increases Jailbreak Risk for Large Language Models

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Xiong, Chen, He, Zhiyuan, Chen, Pin-Yu, Ko, Ching-Yun, Ho, Tsung-Yi
Format:	Preprint
Published:	2026
Subjects:	Cryptography and Security Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Token Highlighter: Inspecting and Mitigating Jailbreak Prompts for Large Language Models
by: Hu, Xiaomeng, et al.
Published: (2024)

Defensive Prompt Patch: A Robust and Interpretable Defense of LLMs against Jailbreak Attacks
by: Xiong, Chen, et al.
Published: (2024)

Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes
by: Hu, Xiaomeng, et al.
Published: (2024)

Steering in the Shadows: Causal Amplification for Activation Space Attacks in Large Language Models
by: Xu, Zhiyuan, et al.
Published: (2025)

SafeSteer: Adaptive Subspace Steering for Efficient Jailbreak Defense in Vision-Language Models
by: Zeng, Xiyu, et al.
Published: (2025)

ShallowJail: Steering Jailbreaks against Large Language Models
by: Liu, Shang, et al.
Published: (2026)

Attention Slipping: A Mechanistic Understanding of Jailbreak Attacks and Defenses in LLMs
by: Hu, Xiaomeng, et al.
Published: (2025)

PII Jailbreaking in LLMs via Activation Steering Reveals Personal Information Leakage
by: Nakka, Krishna Kanth, et al.
Published: (2025)

Adaptive Probe-based Steering for Robust LLM Jailbreaking
by: Chen, Junxi, et al.
Published: (2026)

Prefix Guidance: A Steering Wheel for Large Language Models to Defend Against Jailbreak Attacks
by: Zhao, Jiawei, et al.
Published: (2024)

Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion
by: Pramanik, Vishal, et al.
Published: (2026)

Sugar-Coated Poison: Benign Generation Unlocks LLM Jailbreaking
by: Wu, Yu-Hang, et al.
Published: (2025)

SARSteer: Safeguarding Large Audio-Language Models via Safe-Ablated Refusal Steering
by: Lin, Weilin, et al.
Published: (2025)

Ellipsoid Control: A White-list Jailbreak Defense via Benign Latent Modeling
by: Chen, Luoyu, et al.
Published: (2026)

Malicious and Unintentional Disclosure Risks in Large Language Models for Code Generation
by: Rabin, Rafiqul, et al.
Published: (2025)

Amnesia: Adversarial Semantic Layer Specific Activation Steering in Large Language Models
by: Raza, Ali, et al.
Published: (2026)

SteerDiff: Steering towards Safe Text-to-Image Diffusion Models
by: Zhang, Hongxiang, et al.
Published: (2024)

Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation
by: Chen, Luoyu, et al.
Published: (2026)

AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender
by: Zhao, Weixiang, et al.
Published: (2025)

Trojan Activation Attack: Red-Teaming Large Language Models using Activation Steering for Safety-Alignment
by: Wang, Haoran, et al.
Published: (2023)

Unleashing the Unseen: Harnessing Benign Datasets for Jailbreaking Large Language Models
by: Zhao, Wei, et al.
Published: (2024)

SQL Injection Jailbreak: A Structural Disaster of Large Language Models
by: Zhao, Jiawei, et al.
Published: (2024)

MASCing: Configurable Mixture-of-Experts Behavior via Activation Steering Masks
by: Lintelo, Jona te, et al.
Published: (2026)

Distract Large Language Models for Automatic Jailbreak Attack
by: Xiao, Zeguan, et al.
Published: (2024)

VillanDiffusion: A Unified Backdoor Attack Framework for Diffusion Models
by: Chou, Sheng-Yen, et al.
Published: (2023)

Hide Your Malicious Goal Into Benign Narratives: Jailbreak Large Language Models through Carrier Articles
by: Wang, Zhilong, et al.
Published: (2024)

AutoJailbreak: Exploring Jailbreak Attacks and Defenses through a Dependency Lens
by: Lu, Lin, et al.
Published: (2024)

External Data Extraction Attacks against Retrieval-Augmented Large Language Models
by: He, Yu, et al.
Published: (2025)

Hidden You Malicious Goal Into Benign Narratives: Jailbreak Large Language Models through Logic Chain Injection
by: Wang, Zhilong, et al.
Published: (2024)

Don't Listen To Me: Understanding and Exploring Jailbreak Prompts of Large Language Models
by: Yu, Zhiyuan, et al.
Published: (2024)

Steering Dialogue Dynamics for Robustness against Multi-turn Jailbreaking Attacks
by: Hu, Hanjiang, et al.
Published: (2025)

Prefill-level Jailbreak: A Black-Box Risk Analysis of Large Language Models
by: Li, Yakai, et al.
Published: (2025)

Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors
by: Yin, Rui, et al.
Published: (2026)

Towards Robust Multimodal Large Language Models Against Jailbreak Attacks
by: Yin, Ziyi, et al.
Published: (2025)

from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors
by: Yan, Yu, et al.
Published: (2025)

JBShield: Defending Large Language Models from Jailbreak Attacks through Activated Concept Analysis and Manipulation
by: Zhang, Shenyi, et al.
Published: (2025)

"Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models
by: Shen, Xinyue, et al.
Published: (2023)

FlowSteer: Prompt-Only Workflow Steering Exposes Planning-Time Vulnerabilities in Multi-Agent LLM Systems
by: Li, Fanxiao, et al.
Published: (2026)

Trust Me, Import This: Dependency Steering Attacks via Malicious Agent Skills
by: Liu, Yiyong, et al.
Published: (2026)

Harnessing Task Overload for Scalable Jailbreak Attacks on Large Language Models
by: Dong, Yiting, et al.
Published: (2024)