Image de couverture de livre

False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalize

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Wang, Cheng, Wei, Zeming, Liu, Qin, Chen, Muhao
Format:	Preprint
Publié:	2025
Sujets:	Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities
par: Mo, Lingbo, et autres
Publié: (2023)

Two Heads are Better than One: Nested PoE for Robust Defense Against Multi-Backdoors
par: Graf, Victoria, et autres
Publié: (2024)

Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting
par: Wang, Cheng, et autres
Publié: (2026)

Securing Multi-turn Conversational Language Models From Distributed Backdoor Triggers
par: Tong, Terry, et autres
Publié: (2024)

Monotonic Paraphrasing Improves Generalization of Language Model Prompting
par: Liu, Qin, et autres
Publié: (2024)

Multilingual Embedding Probes Fail to Generalize Across Learner Corpora
par: Lyngbaek, Laurits, et autres
Publié: (2026)

SudoLM: Learning Access Control of Parametric Knowledge with Authorization Alignment
par: Liu, Qin, et autres
Publié: (2024)

Why Retrieval-Augmented Generation Fails: A Graph Perspective
par: Guo, Kai, et autres
Publié: (2026)

Tracking the Limits of Knowledge Propagation: How LLMs Fail at Multi-Step Reasoning with Conflicting Knowledge
par: Feng, Yiyang, et autres
Publié: (2026)

Why Instruction-Based Unlearning Fails in Diffusion Models?
par: Zhang, Zeliang, et autres
Publié: (2026)

Why AI-Generated Text Detection Fails: Evidence from Explainable AI Beyond Benchmark Accuracy
par: Pudasaini, Shushanta, et autres
Publié: (2026)

Secure LLM Fine-Tuning via Safety-Aware Probing
par: Wu, Chengcan, et autres
Publié: (2025)

Feature-Aware Malicious Output Detection and Mitigation
par: Dong, Weilong, et autres
Publié: (2025)

Unraveling Cross-Modality Knowledge Conflicts in Large Vision-Language Models
par: Zhu, Tinghui, et autres
Publié: (2024)

Accommodation and Epistemic Vigilance: A Pragmatic Account of Why LLMs Fail to Challenge Harmful Beliefs
par: Cheng, Myra, et autres
Publié: (2026)

Why Better Cross-Lingual Alignment Fails for Better Cross-Lingual Transfer: Case of Encoders
par: Veitsman, Yana, et autres
Publié: (2026)

Why Chain of Thought Fails in Clinical Text Understanding
par: Wu, Jiageng, et autres
Publié: (2025)

Why Gaussian Diffusion Models Fail on Discrete Data and How to Prevent It?
par: Shabalin, Alexander, et autres
Publié: (2026)

Detection Is Cheap, Routing Is Learned: Why Refusal-Based Alignment Evaluation Fails
par: Frank, Gregory N.
Publié: (2026)

From Shortcuts to Triggers: Backdoor Defense with Denoised PoE
par: Liu, Qin, et autres
Publié: (2023)

Why Do LLM-based Web Agents Fail? A Hierarchical Planning Perspective
par: Aghzal, Mohamed, et autres
Publié: (2026)

Why LoRA Fails to Forget: Regularized Low-Rank Adaptation Against Backdoors in Language Models
par: Luong, Hoang-Chau, et autres
Publié: (2026)

Familiarity-Aware Evidence Compression for Retrieval-Augmented Generation
par: Jung, Dongwon, et autres
Publié: (2024)

AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models
par: Liu, Xiaogeng, et autres
Publié: (2023)

Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models
par: Xue, Chao, et autres
Publié: (2026)

Why Semantic Entropy Fails: Geometry-Aware and Calibrated Uncertainty for Policy Optimization
par: Zhang, Zheyuan, et autres
Publié: (2026)

"Not Aligned" is Not "Malicious": Being Careful about Hallucinations of Large Language Models' Jailbreak
par: Mei, Lingrui, et autres
Publié: (2024)

Why LLMs Fail at Causal Discovery and How Interventional Agents Escape
par: Roy, Amartya, et autres
Publié: (2026)

Why Do Speech Language Models Fail to Generate Semantically Coherent Outputs? A Modality Evolving Perspective
par: Wang, Hankun, et autres
Publié: (2024)

DebugLM: Learning Traceable Training Data Provenance for LLMs
par: Mo, Wenjie Jacky, et autres
Publié: (2026)

DeepEdit: Knowledge Editing as Decoding with Constraints
par: Wang, Yiwei, et autres
Publié: (2024)

Privacy-Preserving Language Model Inference with Instance Obfuscation
par: Yao, Yixiang, et autres
Publié: (2024)

The Mirage of Performance Gains: Why Contrastive Decoding Fails to Mitigate Object Hallucinations in MLLMs?
par: Yin, Hao, et autres
Publié: (2025)

Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations
par: Agarwal, Ananth, et autres
Publié: (2025)

Why Bonds Fail Differently? Explainable Multimodal Learning for Multi-Class Default Prediction
par: Lu, Yi, et autres
Publié: (2025)

The False Resonance: A Critical Examination of Emotion Embedding Similarity for Speech Generation Evaluation
par: Tsai, Yun-Shao, et autres
Publié: (2026)

Semantic and Contextual Modeling for Malicious Comment Detection with BERT-BiLSTM
par: Fang, Zhou, et autres
Publié: (2025)

On the Risk of Evidence Pollution for Malicious Social Text Detection in the Era of LLMs
par: Wan, Herun, et autres
Publié: (2024)

More Rounds, More Noise: Why Multi-Turn Review Fails to Improve Cross-Context Verification
par: Tae-Eun, Song
Publié: (2026)

Large Language Models are Skeptics: False Negative Problem of Input-conflicting Hallucination
par: Song, Jongyoon, et autres
Publié: (2024)