Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Internalizing Safety Understanding in Large Reasoning Models via Verification

About

While explicit Chain-of-Thought (CoT) empowers large reasoning models (LRMs), it enables the generation of riskier final answers. Current alignment paradigms primarily rely on externally enforced compliance, optimizing models to detect malicious prompts rather than evaluating the safety of their own outputs. We argue that this approach remains largely behavioral: our empirical analysis reveals that ostensibly aligned models lack intrinsic safety understanding, often failing to verify their own response safety and remaining vulnerable to adversarial jailbreaks. To address this fundamental limitation, we propose Safety Internal (SInternal), a framework that internalizes safety specifications by training LRMs exclusively on safety verification tasks to critique their own generated answers using expert reasoning trajectories. We demonstrate that learning to verify induces a strong generalization for response safety, significantly enhancing robustness against out-of-domain jailbreaks. Furthermore, when combined with reinforcement learning, SInternal serves as a superior initialization compared to standard supervised fine-tuning, suggesting that internalizing safety understanding creates a more robust foundation for alignment than merely mimicking safe behaviors. Our codes are available at https://github.com/AlphaLab-USTC/SInternal

Yi Zhang, Yuxin Chen, Leheng Sheng, Dongcheng Zhang, Chaochao Lu, Xiang Wang, An Zhang• 2026

Related benchmarks

TaskDatasetResultRank
Over-refusalXSTest--
102
Safety EvaluationStrongREJECT
Attack Success Rate0.6
77
Safety EvaluationWildJailbreak
ASR0.068
70
ReasoningMATH
Success Score (MATH)89.8
15
Harmful Content DetectionFORTRESS
ASR18.6
12
Jailbreak evaluationJailBreak-R1
Attack Success Rate (ASR)1.3
12
ReasoningAIME
Reasoning Accuracy83.3
12
Safety EvaluationFORTRESS
JailBreak Score2.8
12
General ReasoningMATH
pass@189.6
12
General ReasoningAIME 2024
pass@1686.7
12
Showing 10 of 17 rows

Other info

Follow for update