Cover Image

LLM Safety From Within: Detecting Harmful Content with Internal Representations

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Jiao, Difan, Liu, Yilun, Yuan, Ye, Tang, Zhenwei, Du, Linfeng, Wu, Haolun, Anderson, Ashton
Format:	Preprint
Published:	2026
Subjects:	Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

SPIN: Sparsifying and Integrating Internal Neurons in Large Language Models for Text Classification
by: Jiao, Difan, et al.
Published: (2023)

SEAM: Semantically Equivalent Across Modalities Benchmark for Vision-Language Models
by: Tang, Zhenwei, et al.
Published: (2025)

ThinkTwice: Jointly Optimizing Large Language Models for Reasoning and Self-Refinement
by: Jiao, Difan, et al.
Published: (2026)

Maia-2: A Unified Model for Human-AI Alignment in Chess
by: Tang, Zhenwei, et al.
Published: (2024)

Learning to Imitate with Less: Efficient Individual Behavior Modeling in Chess
by: Tang, Zhenwei, et al.
Published: (2025)

ChessQA: Evaluating Large Language Models for Chess Understanding
by: Wen, Qianfeng, et al.
Published: (2025)

Level Up: Defining and Exploiting Transitional Problems for Curriculum Learning
by: Tang, Zhenwei, et al.
Published: (2026)

MINER: Mining Multimodal Internal Representation for Efficient Retrieval
by: Li, Weien, et al.
Published: (2026)

Grounded Chess Reasoning in Language Models via Master Distillation
by: Tang, Zhenwei, et al.
Published: (2026)

Understanding LLM Behavior When Encountering User-Supplied Harmful Content in Harmless Tasks
by: Chu, Junjie, et al.
Published: (2026)

Harmful Suicide Content Detection
by: Park, Kyumin, et al.
Published: (2024)

MV-Debate: Multi-view Agent Debate with Dynamic Reflection Gating for Multimodal Harmful Content Detection in Social Media
by: Lu, Rui, et al.
Published: (2025)

ChineseHarm-Bench: A Chinese Harmful Content Detection Benchmark
by: Liu, Kangwei, et al.
Published: (2025)

Beyond Accuracy: An Explainability-Driven Analysis of Harmful Content Detection
by: Dhara, Trishita, et al.
Published: (2026)

TEARS: Textual Representations for Scrutable Recommendations
by: Penaloza, Emiliano, et al.
Published: (2024)

From Internal Representations to Text Quality: A Geometric Approach to LLM Evaluation
by: Yusupov, Viacheslav, et al.
Published: (2025)

Understanding the Dynamics of Demonstration Conflict in In-Context Learning
by: Jiao, Difan, et al.
Published: (2026)

Prefix Probing: Lightweight Harmful Content Detection for Large Language Models
by: Yang, Jirui, et al.
Published: (2025)

Probing LLM Hallucination from Within: Perturbation-Driven Approach via Internal Knowledge
by: Lee, Seongmin, et al.
Published: (2024)

Scaling Tasks, Not Samples: Mastering Humanoid Control through Multi-Task Model-Based Reinforcement Learning
by: Liu, Shaohuai, et al.
Published: (2026)

Understanding the Generalization of Stochastic Gradient Adam in Learning Neural Networks
by: Tang, Xuan, et al.
Published: (2025)

RouteGuard: Internal-Signal Detection of Skill Poisoning in LLM Agents
by: Xiao, Wenjie, et al.
Published: (2026)

More of the Same: Persistent Representational Harms Under Increased Representation
by: Mickel, Jennifer, et al.
Published: (2025)

From Training-Free to Adaptive: Empirical Insights into MLLMs' Understanding of Detection Information
by: Jiao, Qirui, et al.
Published: (2024)

When Harmful Content Gets Camouflaged: Unveiling Perception Failure of LVLMs with CamHarmTI
by: Li, Yanhui, et al.
Published: (2025)

Look Before You Leap: Enhancing Attention and Vigilance Regarding Harmful Content with GuidelineLLM
by: Zhang, Shaoqing, et al.
Published: (2024)

Why Do Large Language Models Generate Harmful Content?
by: Ganguli, Rajesh, et al.
Published: (2026)

Opir: Efficient Multi-Task Safety Classification for Toxicity, Jailbreaks, Hate Speech, and Harmful Content
by: Stepanov, Ihor, et al.
Published: (2026)

ChemSafetyBench: Benchmarking LLM Safety on Chemistry Domain
by: Zhao, Haochen, et al.
Published: (2024)

Harmful Visual Content Manipulation Matters in Misinformation Detection Under Multimedia Scenarios
by: Wang, Bing, et al.
Published: (2026)

Beyond Message Passing: A Semantic View of Agent Communication Protocols
by: Yuan, Dun, et al.
Published: (2026)

Internalizing Curriculum Judgment for LLM Reinforcement Fine-Tuning
by: Zheng, Han, et al.
Published: (2026)

OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences
by: Wen, Ming, et al.
Published: (2026)

Internal Representations as Indicators of Hallucinations in Agent Tool Selection
by: Healy, Kait, et al.
Published: (2026)

HarmMetric Eval: Benchmarking Metrics and Judges for LLM Harmfulness Assessment
by: Yang, Langqi, et al.
Published: (2025)

Self-HarmLLM: Can Large Language Model Harm Itself?
by: Kim, Heehwan, et al.
Published: (2025)

A Convergence Analysis of Adaptive Optimizers under Floating-point Quantization
by: Tang, Xuan, et al.
Published: (2025)

DarkPatterns-LLM: A Multi-Layer Benchmark for Detecting Manipulative and Harmful AI Behavior
by: Asif, Sadia, et al.
Published: (2025)

Language Models Exhibit Inconsistent Biases Towards Algorithmic Agents and Human Experts
by: Bo, Jessica Y., et al.
Published: (2026)

Turning Internal Gap into Self-Improvement: Promoting the Generation-Understanding Unification in MLLMs
by: Han, Yujin, et al.
Published: (2025)