Imagen de Portada

LLMs Encode Harmfulness and Refusal Separately

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Zhao, Jiachen, Huang, Jing, Wu, Zhengxuan, Bau, David, Shi, Weiyan
Formato:	Preprint
Publicado:	2025
Materias:	Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Should LLM Safety Be More Than Refusing Harmful Instructions?
por: Maskey, Utsav, et al.
Publicado: (2025)

Beyond Over-Refusal: Scenario-Based Diagnostics and Post-Hoc Mitigation for Exaggerated Refusals in LLMs
por: Yuan, Shuzhou, et al.
Publicado: (2025)

Harmful Prompt Laundering: Jailbreaking LLMs with Abductive Styles and Symbolic Encoding
por: Joo, Seongho, et al.
Publicado: (2025)

Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language Models
por: An, Bang, et al.
Publicado: (2024)

Refusal in LLMs is an Affine Function
por: Marshall, Thomas, et al.
Publicado: (2024)

Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
por: Yuan, Youliang, et al.
Publicado: (2024)

Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs
por: Maskey, Utsav, et al.
Publicado: (2026)

Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behavior
por: Si, Shengyun, et al.
Publicado: (2025)

RAID: Refusal-Aware and Integrated Decoding for Jailbreaking LLMs
por: Nguyen, Tuan T., et al.
Publicado: (2025)

Can Editing LLMs Inject Harm?
por: Chen, Canyu, et al.
Publicado: (2024)

HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
por: Mazeika, Mantas, et al.
Publicado: (2024)

Can LLMs Refuse Questions They Do Not Know? Measuring Knowledge-Aware Refusal in Factual Tasks
por: Pan, Wenbo, et al.
Publicado: (2025)

RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations
por: Huang, Jing, et al.
Publicado: (2024)

DataPuzzle: Breaking Free from the Hallucinated Promise of LLMs in Data Analysis
por: Zhang, Zhengxuan, et al.
Publicado: (2025)

Oolong: Investigating What Makes Transfer Learning Hard with Controlled Studies
por: Wu, Zhengxuan, et al.
Publicado: (2022)

Self and Cross-Model Distillation for LLMs: Effective Methods for Refusal Pattern Alignment
por: Li, Jie, et al.
Publicado: (2024)

Silenced Biases: The Dark Side LLMs Learned to Refuse
por: Himelstein, Rom, et al.
Publicado: (2025)

Learning to Refuse: Towards Mitigating Privacy Risks in LLMs
por: Liu, Zhenhua, et al.
Publicado: (2024)

Safety-Aligned Weights Are Not Enough: Refusal-Teacher-Guided Finetuning Enhances Safety and Downstream Performance under Harmful Finetuning Attacks
por: Ham, Seokil, et al.
Publicado: (2025)

Cannot or Should Not? Automatic Analysis of Refusal Composition in IFT/RLHF Datasets and Refusal Behavior of Black-Box LLMs
por: von Recum, Alexander, et al.
Publicado: (2024)

Do Prevalent Bias Metrics Capture Allocational Harms from LLMs?
por: Cyberey, Hannah, et al.
Publicado: (2024)

Can LLMs Rank the Harmfulness of Smaller LLMs? We are Not There Yet
por: Atil, Berk, et al.
Publicado: (2025)

Elucidating Mechanisms of Demographic Bias in LLMs for Healthcare
por: Ahsan, Hiba, et al.
Publicado: (2025)

Measuring and Enhancing Trustworthiness of LLMs in RAG through Grounded Attributions and Learning to Refuse
por: Song, Maojia, et al.
Publicado: (2024)

ReCOGS: How Incidental Details of a Logical Form Overshadow an Evaluation of Semantic Interpretation
por: Wu, Zhengxuan, et al.
Publicado: (2023)

Do Reasoning LLMs Refuse What They Infer in Long Contexts?
por: Fu, Yu, et al.
Publicado: (2026)

Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics
por: García-Ferrero, Iker, et al.
Publicado: (2025)

How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs
por: Zeng, Yi, et al.
Publicado: (2024)

Token Erasure as a Footprint of Implicit Vocabulary Items in LLMs
por: Feucht, Sheridan, et al.
Publicado: (2024)

SafeConstellations: Mitigating Over-Refusals in LLMs Through Task-Aware Representation Steering
por: Maskey, Utsav, et al.
Publicado: (2025)

Does Refusal Training in LLMs Generalize to the Past Tense?
por: Andriushchenko, Maksym, et al.
Publicado: (2024)

Don't Say No: Jailbreaking LLM by Suppressing Refusal
por: Zhou, Yukai, et al.
Publicado: (2024)

Safety Alignment of Large Language Models via Contrasting Safe and Harmful Distributions
por: Zhang, Xiaoyun, et al.
Publicado: (2024)

Bridging Cognition and Emotion: Empathy-Driven Multimodal Misinformation Detection
por: Wang, Zihan, et al.
Publicado: (2025)

Unintended Harms of Value-Aligned LLMs: Psychological and Empirical Insights
por: Choi, Sooyung, et al.
Publicado: (2025)

Jailbreaking Commercial Black-Box LLMs with Explicitly Harmful Prompts
por: Zhang, Chiyu, et al.
Publicado: (2025)

PluriHarms: Benchmarking the Full Spectrum of Human Judgments on AI Harm
por: Li, Jing-Jing, et al.
Publicado: (2026)

Learning-Time Encoding Shapes Unlearning in LLMs
por: Wu, Ruihan, et al.
Publicado: (2025)

WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs
por: Han, Seungju, et al.
Publicado: (2024)

Read as You See: Guiding Unimodal LLMs for Low-Resource Explainable Harmful Meme Detection
por: Pan, Fengjun, et al.
Publicado: (2025)