SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908578584985600 |
|---|---|
| author | Feng, Kehua Ding, Keyan Wang, Yuhao Li, Menghan Wei, Fanjunduo Wang, Xinda Zhang, Qiang Chen, Huajun |
| author_facet | Feng, Kehua Ding, Keyan Wang, Yuhao Li, Menghan Wei, Fanjunduo Wang, Xinda Zhang, Qiang Chen, Huajun |
| contents | Recent advancements in large language models (LLMs) have accelerated progress toward artificial general intelligence, yet their potential to generate harmful content poses critical safety challenges. Existing alignment methods often struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. In this work, we propose SAFER, a framework for Safety Alignment via eFficient Ex-Ante Reasoning. Our approach instantiates structured Ex-Ante reasoning through initial assessment, rule verification, and path calibration, and embeds predefined safety rules to provide transparent and verifiable safety judgments. Specifically, our approach consists of two training stages: (1) supervised fine-tuning with synthetic traces to teach the multi-stage Ex-Ante reasoning, and (2) step-level reasoning preference optimization to jointly enhance safety, utility, and efficiency. Experiments on multiple open-source LLMs demonstrate that SAFER significantly enhances safety performance while maintaining helpfulness and response efficiency. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_02725 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning Feng, Kehua Ding, Keyan Wang, Yuhao Li, Menghan Wei, Fanjunduo Wang, Xinda Zhang, Qiang Chen, Huajun Computation and Language Recent advancements in large language models (LLMs) have accelerated progress toward artificial general intelligence, yet their potential to generate harmful content poses critical safety challenges. Existing alignment methods often struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. In this work, we propose SAFER, a framework for Safety Alignment via eFficient Ex-Ante Reasoning. Our approach instantiates structured Ex-Ante reasoning through initial assessment, rule verification, and path calibration, and embeds predefined safety rules to provide transparent and verifiable safety judgments. Specifically, our approach consists of two training stages: (1) supervised fine-tuning with synthetic traces to teach the multi-stage Ex-Ante reasoning, and (2) step-level reasoning preference optimization to jointly enhance safety, utility, and efficiency. Experiments on multiple open-source LLMs demonstrate that SAFER significantly enhances safety performance while maintaining helpfulness and response efficiency. |
| title | SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.02725 |