Skip to content
conftrace_
2025 ICML ICML 2025

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

Abstract

No abstract available. No abstract was captured for this paper when its proceedings were scraped - most often an older edition published without one online. How the corpus is built.

Authors

Zihan Guan, Mengxuan Hu, Ronghang Zhu, Sheng Li, Anil Vullikanti

Topics

Artificial Intelligence > Core AI > AI Safety Machine Learning > Learning Types > Transfer Learning Artificial Intelligence > Core AI > Large Language Models

Keywords

fine tuning large language model safety degradation alignment safeguards outlier samples self-inf-n
Download PDF