Continual Safety Alignment via Gradient-Based Sample Selection

April 19, 2026 ยท Grace Period ยท ๐Ÿ› ACL 2026 (Findings)

โณ Grace Period
This paper is less than 90 days old. We give authors time to release their code before passing judgment.
Authors Thong Bach, Dung Nguyen, Thao Minh Le, Truyen Tran arXiv ID 2604.17215 Category cs.LG: Machine Learning Citations 0 Venue ACL 2026 (Findings)
Abstract
Large language models require continuous adaptation to new tasks while preserving safety alignment. However, fine-tuning on even benign data often compromises safety behaviors, including refusal of harmful requests, truthfulness, and commonsense reasoning. We investigate which training samples cause alignment drift through a data-centric lens. Our empirical analysis shows samples contribute unequally: high-gradient samples cause greater safety degradation and drive models toward pretrained distributions, while moderate-gradient samples enable task learning with minimal alignment loss. We propose gradient-based sample selection that filters high-gradient samples during fine-tuning. Across multiple model families on continual domain tasks, our method substantially improves alignment preservation while maintaining competitive task performance, without requiring curated safe data or architectural modifications. Our method is robust across selection ratios, task orderings, and diverse attack benchmarks.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

๐Ÿ“œ Similar Papers

In the same crypt โ€” Machine Learning