Cover Image

Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safety

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Guan, Zihan, Hu, Mengxuan, Zhu, Ronghang, Li, Sheng, Vullikanti, Anil
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

UFID: A Unified Framework for Input-level Backdoor Detection on Diffusion Models
by: Guan, Zihan, et al.
Published: (2024)

Are LLMs Ready for Neural-integrated Mechanistic Modeling? A Benchmark and Agentic Framework
by: Guan, Zihan, et al.
Published: (2026)

Outlier-weighed Layerwise Sampling for LLM Fine-tuning
by: Li, Pengxiang, et al.
Published: (2024)

Large Language Models Lack Temporal Awareness of Medical Knowledge
by: Guan, Zihan, et al.
Published: (2026)

What is in Your Safe Data? Identifying Benign Data that Breaks Safety
by: He, Luxi, et al.
Published: (2024)

What Makes and Breaks Safety Fine-tuning? A Mechanistic Study
by: Jain, Samyak, et al.
Published: (2024)

Improving Sampling Methods for Fine-tuning SentenceBERT in Text Streams
by: Garcia, Cristiano Mesquita, et al.
Published: (2024)

No Free Lunch: Retrieval-Augmented Generation Undermines Fairness in LLMs, Even for Vigilant Users
by: Hu, Mengxuan, et al.
Published: (2024)

Memento: Fine-tuning LLM Agents without Fine-tuning LLMs
by: Zhou, Huichi, et al.
Published: (2025)

GRASS: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning
by: Tian, Kaiyuan, et al.
Published: (2026)

Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling
by: Huang, Zeyu, et al.
Published: (2025)

Layer-wise Importance Matters: Less Memory for Better Performance in Parameter-efficient Fine-tuning of Large Language Models
by: Yao, Kai, et al.
Published: (2024)

Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning
by: Wang, Guoli, et al.
Published: (2026)

RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization
by: Huang, Xijie, et al.
Published: (2024)

Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets
by: Hsiung, Lei, et al.
Published: (2025)

Multitask Mayhem: Unveiling and Mitigating Safety Gaps in LLMs Fine-tuning
by: Jan, Essa, et al.
Published: (2024)

Privacy Collapse: Benign Fine-Tuning Can Break Contextual Privacy in Language Models
by: Goel, Anmol, et al.
Published: (2026)

Advancing Parameter Efficiency in Fine-tuning via Representation Editing
by: Wu, Muling, et al.
Published: (2024)

Objective Matters: Fine-Tuning Objectives Shape Safety, Robustness, and Persona Drift
by: Vennemeyer, Daniel, et al.
Published: (2026)

PMSS: Pretrained Matrices Skeleton Selection for LLM Fine-tuning
by: Wang, Qibin, et al.
Published: (2024)

SEE: Continual Fine-tuning with Sequential Ensemble of Experts
by: Wang, Zhilin, et al.
Published: (2025)

Fundamental Safety-Capability Trade-offs in Fine-tuning Large Language Models
by: Chen, Pin-Yu, et al.
Published: (2025)

CFSafety: Comprehensive Fine-grained Safety Assessment for LLMs
by: Liu, Zhihao, et al.
Published: (2024)

Learning Rate Matters: Vanilla LoRA May Suffice for LLM Fine-tuning
by: Lee, Yu-Ang, et al.
Published: (2026)

Efficient Ensemble for Fine-tuning Language Models on Multiple Datasets
by: Li, Dongyue, et al.
Published: (2025)

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning
by: Perin, Gabriel J., et al.
Published: (2025)

A Classical View on Benign Overfitting: The Role of Sample Size
by: Park, Junhyung, et al.
Published: (2025)

SEAL: Safety-enhanced Aligned LLM Fine-tuning via Bilevel Data Selection
by: Shen, Han, et al.
Published: (2024)

REAL Sampling: Boosting Factuality and Diversity of Open-Ended Generation via Asymptotic Entropy
by: Chang, Haw-Shiuan, et al.
Published: (2024)

Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack
by: Huang, Tiansheng, et al.
Published: (2024)

CoMoE: Contrastive Representation for Mixture-of-Experts in Parameter-Efficient Fine-tuning
by: Feng, Jinyuan, et al.
Published: (2025)

Topic Modeling with Fine-tuning LLMs and Bag of Sentences
by: Schneider, Johannes
Published: (2024)

On the Loss of Context-awareness in General Instruction Fine-tuning
by: Wang, Yihan, et al.
Published: (2024)

DataShield: Safety-degrading Data Filtering for LLM Benign Instruction Fine-Tuning
by: Zhang, Junbo, et al.
Published: (2026)

Learn Hard Problems During RL with Reference Guided Fine-tuning
by: Wu, Yangzhen, et al.
Published: (2026)

Threshold Filtering Packing for Supervised Fine-Tuning: Training Related Samples within Packs
by: Dong, Jiancheng, et al.
Published: (2024)

Two-stage LLM Fine-tuning with Less Specialization and More Generalization
by: Wang, Yihan, et al.
Published: (2022)

A General Framework to Enhance Fine-tuning-based LLM Unlearning
by: Ren, Jie, et al.
Published: (2025)

ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning
by: Lin, Zihan, et al.
Published: (2026)

Dissecting Outlier Dynamics in LLM NVFP4 Pretraining
by: Dong, Peijie, et al.
Published: (2026)