Arrow Research search
Back to ICML

ICML 2025

Safety Reasoning with Guidelines

Conference Paper Accept (poster) Artificial Intelligence ยท Machine Learning

Abstract

Training safe LLMs remains a critical challenge. The most widely used method, Refusal Training (RT), struggles to generalize against various Out-of-Distribution (OOD) jailbreaking attacks. Although various advanced methods have been proposed to address this issue, we instead question whether OOD attacks inherently surpass the capability of vanilla RT. Evaluations using Best-of-N (BoN) reveal significant safety improvements as N increases, indicating models possess adequate latent safety knowledge but RT fails to consistently elicit it under OOD scenarios. Further domain adaptation analysis reveals that direct RT causes reliance on superficial shortcuts, resulting in non-generalizable representation mappings. Inspired by our findings, we propose training model to perform safety reasoning for each query. Specifically, we synthesize reasoning supervision aligned with specified guidelines that reflect diverse perspectives on safety knowledge. This encourages model to engage in deeper reasoning, explicitly eliciting and utilizing latent safety knowledge for each query. Extensive experiments show that our method significantly improves model generalization against OOD attacks.

Authors

Keywords

  • Safety Alignment
  • Safety Reasoning
  • Safety Generalization
  • OOD Generalization

Context

Venue
International Conference on Machine Learning
Archive span
1993-2025
Indexed papers
16471
Paper id
613505795348115228
v2026.09.13