Image de couverture de livre

Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Du, Yanrui, Fan, Fenglei, Zhao, Sendong, Cao, Jiawei, Lin, Qika, He, Kai, Liu, Ting, Qin, Bing, Feng, Mengling
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

Toward Secure Tuning: Mitigating Security Risks from Instruction Fine-Tuning
par: Du, Yanrui, et autres
Publié: (2024)

MoGU V2: Toward a Higher Pareto Frontier Between Model Usability and Security
par: Du, Yanrui, et autres
Publié: (2025)

From Latent Signals to Reflection Behavior: Tracing Meta-Cognitive Activation Trajectory in R1-Style LLMs
par: Du, Yanrui, et autres
Publié: (2026)

S3-CoT: Self-Sampled Succinct Reasoning Enables Efficient Chain-of-Thought LLMs
par: Du, Yanrui, et autres
Publié: (2026)

Don't Ignore Dual Logic Ability of LLMs while Privatizing: A Data-Intensive Analysis in Medical Domain
par: Du, Yanrui, et autres
Publié: (2023)

SL-BiLEM: Structured Learnable Behavior-in-the-Loop Epidemic Modeling for Forecasting and Policy Evaluation
par: Wang, Haochun, et autres
Publié: (2026)

MolTailor: Tailoring Chemical Molecular Representation to Specific Tasks via Text Prompts
par: Guo, Haoqiang, et autres
Publié: (2024)

Analyzing the Inherent Response Tendency of LLMs: Real-World Instructions-Driven Jailbreak
par: Du, Yanrui, et autres
Publié: (2023)

Beyond Semantics: An Evidential Reasoning-Aware Multi-View Learning Framework for Trustworthy Mental Health Prediction
par: Ruan, Yucheng, et autres
Publié: (2026)

MolFusion: Multimodal Fusion Learning for Molecular Representations via Multi-granularity Views
par: Cai, Muzhen, et autres
Publié: (2024)

ST-USleepNet: A Spatial-Temporal Coupling Prominence Network for Multi-Channel Sleep Staging
par: Ma, Jingying, et autres
Publié: (2024)

MoGU: A Framework for Enhancing Safety of Open-Sourced LLMs While Preserving Their Usability
par: Du, Yanrui, et autres
Publié: (2024)

From Artificially Real to Real: Leveraging Pseudo Data from Large Language Models for Low-Resource Molecule Discovery
par: Chen, Yuhan, et autres
Publié: (2023)

OptiSet: Unified Optimizing Set Selection and Ranking for Retrieval-Augmented Generation
par: Jiang, Yi, et autres
Publié: (2026)

A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics
par: He, Kai, et autres
Publié: (2023)

EsurvFusion: An evidential multimodal survival fusion model based on Gaussian random fuzzy numbers
par: Huang, Ling, et autres
Publié: (2024)

Beyond Direct Diagnosis: LLM-based Multi-Specialist Agent Consultation for Automatic Diagnosis
par: Wang, Haochun, et autres
Publié: (2024)

Towards Unified Neurosymbolic Reasoning on Knowledge Graphs
par: Lin, Qika, et autres
Publié: (2025)

Self-supervised Quantized Representation for Seamlessly Integrating Knowledge Graphs with Large Language Models
par: Lin, Qika, et autres
Publié: (2025)

A Survey of Quantized Graph Representation Learning: Connecting Graph Structures with Large Language Models
par: Lin, Qika, et autres
Publié: (2025)

Manifold-based Verbalizer Space Re-embedding for Tuning-free Prompt-based Classification
par: Wang, Haochun, et autres
Publié: (2023)

Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs
par: Zhao, Weixiang, et autres
Publié: (2025)

GRAIT: Gradient-Driven Refusal-Aware Instruction Tuning for Effective Hallucination Mitigation
par: Zhu, Runchuan, et autres
Publié: (2025)

Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behavior
par: Si, Shengyun, et autres
Publié: (2025)

Uncovering the Role of Initial Saliency in U-Shaped Attention Bias: Scaling Initial Token Weight for Enhanced Long-Text Processing
par: Qiang, Zewen, et autres
Publié: (2025)

CodeBrain: Bridging Decoupled Tokenizer and Multi-Scale Architecture for EEG Foundation Model
par: Ma, Jingying, et autres
Publié: (2025)

Has Multimodal Learning Delivered Universal Intelligence in Healthcare? A Comprehensive Survey
par: Lin, Qika, et autres
Publié: (2024)

Refusal Direction is Universal Across Safety-Aligned Languages
par: Wang, Xinpeng, et autres
Publié: (2025)

Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment
par: Xue, Zhiyu, et autres
Publié: (2026)

GSEM: Graph-based Self-Evolving Memory for Experience Augmented Clinical Reasoning
par: Han, Xiao, et autres
Publié: (2026)

Learning to Refuse: Towards Mitigating Privacy Risks in LLMs
par: Liu, Zhenhua, et autres
Publié: (2024)

Toward a Multi-View Brain Network Foundation Model: Cross-View Consistency Learning Across Arbitrary Atlases
par: Xu, Jiaxing, et autres
Publié: (2026)

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

medR: Reward Engineering for Clinical Offline Reinforcement Learning via Tri-Drive Potential Functions
par: Xu, Qianyi, et autres
Publié: (2026)

SCOPE: Structured Prototype-Guided Adaptation for EEG Foundation Models with Limited Labels
par: Ma, Jingying, et autres
Publié: (2026)

Knowledge-tuning Large Language Models with Structured Medical Knowledge Bases for Reliable Response Generation in Chinese
par: Wang, Haochun, et autres
Publié: (2023)

AS-ES Learning: Towards Efficient CoT Learning in Small Models
par: Xi, Nuwa, et autres
Publié: (2024)

Beyond Frameworks: Unpacking Collaboration Strategies in Multi-Agent Systems
par: Wang, Haochun, et autres
Publié: (2025)

LLMs May Perform MCQA by Selecting the Least Incorrect Option
par: Wang, Haochun, et autres
Publié: (2024)

DualEdit: Mitigating Safety Fallback in LLM Backdoor Editing via Affirmation-Refusal Regulation
par: Jiang, Houcheng, et autres
Publié: (2025)