mlsanityData to 14 Oct 2026

UnsafeChain: Enhancing Reasoning Model Safety via Hard Cases

arXiv 2507.21652 ↗First posted 29 Jul 2025cs.CLcited by 7

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more
Abstract (arXiv)

As large reasoning models (LRMs) grow more capable, chain-of-thought (CoT) reasoning introduces new safety challenges. Existing SFT-based safety alignment studies dominantly focused on filtering prompts with safe, high-quality responses, while overlooking hard prompts that always elicit harmful outputs. To fill this gap, we introduce UnsafeChain, a safety alignment dataset constructed from hard prompts with diverse sources, where unsafe completions are identified and explicitly corrected into safe responses. By exposing models to unsafe behaviors and guiding their correction, UnsafeChain enhances safety while preserving general reasoning ability. We fine-tune three LRMs on UnsafeChain and compare them against recent SafeChain and STAR-1 across six out-of-distribution and five in-distribution benchmarks. UnsafeChain consistently outperforms prior datasets, with even a 1K subset matching or surpassing baseline performance, demonstrating the effectiveness and generalizability of correction-based supervision. We release our dataset and code at https://github.com/mbzuai-nlp/UnsafeChain

Key facts. Cited by 7 papers 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 29 of its 35 references that are in the map. First posted on arXiv 29 Jul 2025.

Builds on

29 of 35 references.

Closest ancestors · 14

Same direction, newest first

  1. 2025Safety in Large Reasoning Models: A Survey24 Apr
  2. 2025RealSafe-R1: Safety-Aligned DeepSeek-R1 without Compromising Reasoning Capability14 Apr
  3. 2025SaRO: Enhancing LLM Safety through Reasoning-based Alignment13 Apr
  4. 2025STAR-1: Safer Alignment of Reasoning LLMs with 1K Data2 Apr
  5. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
  6. 2025ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails19 Feb
  7. 2025Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking18 Feb
  8. 2025The Hidden Risks of Large Reasoning Models: A Safety Assessment of R118 Feb
  9. 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities17 Feb
  10. 2025Safety Reasoning with Guidelines6 Feb
  11. 2025STAIR: Improving Safety Alignment with Introspective Reasoning4 Feb
  12. 2025The dark deep side of DeepSeek: Fine-tuning attacks against the safety alignment of CoT-enabled models3 Feb
2 more
  1. 2025GuardReasoner: Towards Reasoning-based LLM Safeguards30 Jan
  2. 2024Deliberative Alignment: Reasoning Enables Safer Language Models20 Dec

Other references · 15

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2023Let's Verify Step by Stepcited by 3,789
  6. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  7. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  8. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  9. 2024A StrongREJECT for Empty Jailbreakscited by 385
  10. 2024Frontier Models are Capable of In-context Schemingcited by 287
  11. 2024WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Modelscited by 223
  12. 2025Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMscited by 172
3 more
  1. 2025Equilibrate RLHF: Towards Balancing Helpfulness-Safety Trade-off in Large Language Modelscited by 22
  2. 2024WildVis: Open Source Visualizer for Million-Scale Chat Logs in the Wildcited by 20
  3. 2025Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacementscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 2; 2026: 5.

7 papers.

  1. 2025AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Modelscited by 4
  2. 2026Towards Safer Large Reasoning Models by Promoting Safety Decision-Making before Chain-of-Thought Generationcited by 2
  3. 2026Self-ReSET: Learning to Self-Recover from Unsafe Reasoning Trajectoriescited by 1
  4. 2025RAJ-PGA: Reasoning-Activated Jailbreak and Principle-Guided Alignment Framework for Large Reasoning Modelscited by 0
  5. 2026How Does Prefix Matter in Reasoning Model Tuning?cited by 0
  6. 2026DataRx: Missingness-Aware Sampling for Safer Large Language Model Task-Specific Fine-Tuningcited by 0
  7. 2026Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusalcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment
  2. 2025SafeRBench: Dissecting the Reasoning Safety of Large Language Models
  3. 2025Towards Safe Reasoning in Large Reasoning Models via Corrective Intervention
  4. 2026Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering
  5. 2025Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models
  6. 2026Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models
  7. 2026TRACE: An Evidence-Grounded Benchmark for Safety Evaluation of Large Reasoning Models
  8. 2026Beyond Content Safety: Real-Time Monitoring for Reasoning Vulnerabilities in Large Language Models