Cover Image

Interpretable LLM Guardrails via Sparse Representation Steering

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	He, Zeqing, Wang, Zhibo, Xu, Huiyu, Lin, Hejun, Zhang, Wenhui, Chu, Zhixuan
Format:	Preprint
Published:	2025
Subjects:	Cryptography and Security Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

JailbreakLens: Interpreting Jailbreak Mechanism in the Lens of Representation and Circuit
by: He, Zeqing, et al.
Published: (2024)

RerouteGuard: Understanding and Mitigating Adversarial Risks for LLM Routing
by: Zhang, Wenhui, et al.
Published: (2026)

TWGuard: A Case Study of LLM Safety Guardrails for Localized Linguistic Contexts
by: Chu, Hua-Rong, et al.
Published: (2026)

Can Small Language Models Reliably Resist Jailbreak Attacks? A Comprehensive Evaluation
by: Zhang, Wenhui, et al.
Published: (2025)

RedAgent: Red Teaming Large Language Models with Context-aware Autonomous Language Agent
by: Xu, Huiyu, et al.
Published: (2024)

OpenGuardrails: A Configurable, Unified, and Scalable Guardrails Platform for Large Language Models
by: Wang, Thomas, et al.
Published: (2025)

Triaging Threats to Specialized Guardrails
by: Mo, Wenjie Jacky, et al.
Published: (2026)

OneShield -- the Next Generation of LLM Guardrails
by: DeLuca, Chad, et al.
Published: (2025)

Bag of Tricks for Subverting Reasoning-based Safety Guardrails
by: Chen, Shuo, et al.
Published: (2025)

Black-Box Guardrail Reverse-engineering Attack
by: Yao, Hongwei, et al.
Published: (2025)

Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Control
by: Cyberey, Hannah, et al.
Published: (2025)

NeuroFilter: Privacy Guardrails for Conversational LLM Agents
by: Das, Saswat, et al.
Published: (2026)

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

Jailbreak Antidote: Runtime Safety-Utility Balance via Sparse Representation Adjustment in Large Language Models
by: Shen, Guobin, et al.
Published: (2024)

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

Proof-of-Guardrail in AI Agents and What (Not) to Trust from It
by: Jin, Xisen, et al.
Published: (2026)

PRISM: Privacy-Aware Routing for Adaptive Cloud-Edge LLM Inference via Semantic Sketch Collaboration
by: Zhan, Junfei, et al.
Published: (2025)

ML-Bench&Guard: Policy-Grounded Multilingual Safety Benchmark and Guardrail for Large Language Models
by: Zhao, Yunhan, et al.
Published: (2026)

Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models
by: Yi, Biao, et al.
Published: (2025)

Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM Safety
by: Zhang, Yuyou, et al.
Published: (2025)

LoopTrap: Termination Poisoning Attacks on LLM Agents
by: Xu, Huiyu, et al.
Published: (2026)

From Retrieval to Reasoning: A Framework for Cyber Threat Intelligence NER with Explicit and Adaptive Instructions
by: Peng, Jiaren, et al.
Published: (2025)

Defending LLM Watermarking Against Spoofing Attacks with Contrastive Representation Learning
by: An, Li, et al.
Published: (2025)

Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignment
by: Rad, Melissa Kazemi, et al.
Published: (2025)

Analysing the Safety Pitfalls of Steering Vectors
by: Li, Yuxiao, et al.
Published: (2026)

SEEP: Training Dynamics Grounds Latent Representation Search for Mitigating Backdoor Poisoning Attacks
by: He, Xuanli, et al.
Published: (2024)

SAEs $\textit{Can}$ Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs
by: Muhamed, Aashiq, et al.
Published: (2025)

Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets
by: Hsiung, Lei, et al.
Published: (2025)

SGuard-v1: Safety Guardrail for Large Language Models
by: Lee, JoonHo, et al.
Published: (2025)

Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs
by: Xing, Wenpeng, et al.
Published: (2025)

RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content
by: Yuan, Zhuowen, et al.
Published: (2024)

Auto-Tuning Safety Guardrails for Black-Box Large Language Models
by: Abdulkadir, Perry
Published: (2025)

"Moralized" Multi-Step Jailbreak Prompts: Black-Box Testing of Guardrails in Large Language Models for Verbal Attacks
by: Wang, Libo
Published: (2024)

MaskSQL: Safeguarding Privacy for LLM-Based Text-to-SQL via Abstraction
by: Abedini, Sepideh, et al.
Published: (2025)

Understanding and Mitigating Over-refusal for Large Language Models via Safety Representation
by: Zhang, Junbo, et al.
Published: (2025)

The Earth is Flat because...: Investigating LLMs' Belief towards Misinformation via Persuasive Conversation
by: Xu, Rongwu, et al.
Published: (2023)

InjecGuard: Benchmarking and Mitigating Over-defense in Prompt Injection Guardrail Models
by: Li, Hao, et al.
Published: (2024)

Sparse Autoencoders are Capable LLM Jailbreak Mitigators
by: Assogba, Yannick, et al.
Published: (2026)

ADVERSA: Measuring Multi-Turn Guardrail Degradation and Judge Reliability in Large Language Models
by: Owiredu-Ashley, Harry
Published: (2026)

SATA: A Paradigm for LLM Jailbreak via Simple Assistive Task Linkage
by: Dong, Xiaoning, et al.
Published: (2024)