Reasoning Structure Matters for Safety Alignment of Reasoning Models
On the map: LLM reasoning → Reasoning for agents, planning and code
also in 2 more
Abstract (arXiv)
Large reasoning models (LRMs) achieve strong performance on complex reasoning tasks but often generate harmful responses to malicious user queries. This paper investigates the underlying cause of these safety risks and shows that the issue lies in the reasoning structure itself. Based on this insight, we claim that effective safety alignment can be achieved by altering the reasoning structure. We propose AltTrain, a simple yet effective post training method that explicitly alters the reasoning structure of LRMs. AltTrain is both practical and generalizable, requiring no complex reinforcement learning (RL) training or reward design, only supervised finetuning (SFT) with a lightweight 1K training examples. Experiments across LRM backbones and model sizes demonstrate strong safety alignment, along with robust generalization across reasoning, QA, summarization, and multilingual setting.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 21 of its 35 references that are in the map. First posted on arXiv 21 Apr 2026.
Builds on
21 of 35 references.
Closest ancestors · 8
Same direction, newest first
- 2025How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study21 May
- 2025STAR-1: Safer Alignment of Reasoning LLMs with 1K Data2 Apr
- 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
- 2025Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking18 Feb
- 2025The Hidden Risks of Large Reasoning Models: A Safety Assessment of R118 Feb
- 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities17 Feb
- 2024OpenAI o1 System Card21 Dec
- 2024Deliberative Alignment: Reasoning Enables Safer Language Models20 Dec
Other references · 13
- 2021Training Verifiers to Solve Math Word Problemscited by 10,109
- 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
- 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
- 2021Evaluating Large Language Models Trained on Codecited by 8,320
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
- 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
- 2025s1: Simple test-time scalingcited by 1,341
- 2016Abstractive Text Summarization Using Sequence-to-Sequence RNNs and Beyondcited by 812
- 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
- 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
- 2023CMMLU: Measuring massive multitask language understanding in Chinesecited by 449
- 2024A StrongREJECT for Empty Jailbreakscited by 385
1 more
- 2024Intention Analysis Makes LLMs A Good Jailbreak Defendercited by 74
Built on by
Citing papers per year: 2026: 1.
1 paper.
- 2026On the Efficiency-Safety Dilemma in Large Reasoning Models20 Sep 2026
Similar, not cited
Close in wording, no citation either way
- 2025Beyond SFT: Reinforcement Learning for Safer Large Reasoning Models with Better Reasoning Ability
- 2026DeShortcut-Align: Decoupling Spurious Shortcuts for Robust Safety Alignment in Large Reasoning Models
- 2026Towards Mitigating Deceptive Safety Alignment in Large Reasoning Models
- 2026THINKSAFE: Self-Generated Safety Alignment for Reasoning Models
- 2026Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models
- 2025Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models
- 2026Towards Safer Large Reasoning Models by Promoting Safety Decision-Making before Chain-of-Thought Generation
- 2025R1-ACT: Efficient Reasoning Model Safety Alignment by Activating Safety Knowledge