Image de couverture de livre

Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak Detection

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Hu, Xulin, Wang, Che, Lim, Wei Yang Bryan, Gao, Jianbo, Chen, Zhong
Format:	Preprint
Publié:	2026
Sujets:	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Refusal Before Decoding: Detecting and Exploiting Refusal Signals in Intermediate LLM Activations
par: Collu, Matteo Gioele, et autres
Publié: (2026)

Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes
par: Hu, Xiaomeng, et autres
Publié: (2024)

From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment
par: Chae, Kyubyung, et autres
Publié: (2025)

A Content-Based Framework for Cybersecurity Refusal Decisions in Large Language Models
par: Linder, Noa, et autres
Publié: (2026)

Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-off
par: Chen, Yu, et autres
Publié: (2026)

Furina: Fragmented Uncertainty-Driven Refusal Instability Attack
par: Wu, Tongxi, et autres
Publié: (2026)

AdapTools: Adaptive Tool-based Indirect Prompt Injection Attacks on Agentic LLMs
par: Wang, Che, et autres
Publié: (2026)

Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?
par: Yin, Qingyu, et autres
Publié: (2025)

Defensive Refusal Bias: How Safety Alignment Fails Cyber Defenders
par: Campbell, David, et autres
Publié: (2026)

Beyond Explicit Refusals: Soft-Failure Attacks on Retrieval-Augmented Generation
par: Zhang, Wentao, et autres
Publié: (2026)

ICON: Indirect Prompt Injection Defense for Agents based on Inference-Time Correction
par: Wang, Che, et autres
Publié: (2026)

Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs
par: Xing, Wenpeng, et autres
Publié: (2025)

Pattern Enhanced Multi-Turn Jailbreaking: Exploiting Structural Vulnerabilities in Large Language Models
par: Nihal, Ragib Amin, et autres
Publié: (2025)

LLM Jailbreak Detection for (Almost) Free!
par: Chen, Guorui, et autres
Publié: (2025)

A Behavioral Fingerprint for Large Language Models: Provenance Tracking via Refusal Vectors
par: Xu, Zhenyu, et autres
Publié: (2026)

Geneshift: Impact of different scenario shift on Jailbreaking LLM
par: Wu, Tianyi, et autres
Publié: (2025)

SeqAR: Jailbreak LLMs with Sequential Auto-Generated Characters
par: Yang, Yan, et autres
Publié: (2024)

Dynamic Adversarial Fine-Tuning Reorganizes Refusal Geometry
par: Lan, Wenhao, et autres
Publié: (2026)

Distract Large Language Models for Automatic Jailbreak Attack
par: Xiao, Zeguan, et autres
Publié: (2024)

RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs
par: Asif, Sadia, et autres
Publié: (2026)

SATA: A Paradigm for LLM Jailbreak via Simple Assistive Task Linkage
par: Dong, Xiaoning, et autres
Publié: (2024)

Exploiting Prefix-Tree in Structured Output Interfaces for Enhancing Jailbreak Attacking
par: Li, Yanzeng, et autres
Publié: (2025)

Attention Slipping: A Mechanistic Understanding of Jailbreak Attacks and Defenses in LLMs
par: Hu, Xiaomeng, et autres
Publié: (2025)

AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint
par: Sheng, Leheng, et autres
Publié: (2025)

Jailbreaking in the Haystack
par: Shah, Rishi Rajesh, et autres
Publié: (2025)

Subtoxic Questions: Dive Into Attitude Change of LLM's Response in Jailbreak Attempts
par: Zhang, Tianyu, et autres
Publié: (2024)

Layer-Level Self-Exposure and Patch: Affirmative Token Mitigation for Jailbreak Attack Defense
par: Ouyang, Yang, et autres
Publié: (2025)

Imperceptible Jailbreaking against Large Language Models
par: Gao, Kuofeng, et autres
Publié: (2025)

Do Reasoning LLMs Refuse What They Infer in Long Contexts?
par: Fu, Yu, et autres
Publié: (2026)

Jailbreaking Multimodal Large Language Models via Shuffle Inconsistency
par: Zhao, Shiji, et autres
Publié: (2025)

RobustKV: Defending Large Language Models against Jailbreak Attacks via KV Eviction
par: Jiang, Tanqiu, et autres
Publié: (2024)

Knowledge-to-Jailbreak: Investigating Knowledge-driven Jailbreaking Attacks for Large Language Models
par: Tu, Shangqing, et autres
Publié: (2024)

JailbreakEval: An Integrated Toolkit for Evaluating Jailbreak Attempts Against Large Language Models
par: Ran, Delong, et autres
Publié: (2024)

MRJ-Agent: An Effective Jailbreak Agent for Multi-Round Dialogue
par: Wang, Fengxiang, et autres
Publié: (2024)

Advancing Jailbreak Strategies: A Hybrid Approach to Exploiting LLM Vulnerabilities and Bypassing Modern Defenses
par: Ahmed, Mohamed, et autres
Publié: (2025)

STShield: Single-Token Sentinel for Real-Time Jailbreak Detection in Large Language Models
par: Wang, Xunguang, et autres
Publié: (2025)

Unleashing the Unseen: Harnessing Benign Datasets for Jailbreaking Large Language Models
par: Zhao, Wei, et autres
Publié: (2024)

How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation
par: Long, Zhuohang, et autres
Publié: (2025)

GPT-4 Jailbreaks Itself with Near-Perfect Success Using Self-Explanation
par: Ramesh, Govind, et autres
Publié: (2024)

Is the System Message Really Important to Jailbreaks in Large Language Models?
par: Zou, Xiaotian, et autres
Publié: (2024)