mlsanityData to 7 Oct 2026

ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails

arXiv 2502.13458 ↗First posted 19 Feb 2025cs.CL cs.AI cs.CRcited by 29

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

also in 1 more
Abstract (arXiv)

Ensuring the safety of large language models (LLMs) is critical as they are deployed in real-world applications. Existing guardrails rely on rule-based filtering or single-pass classification, limiting their ability to handle nuanced safety violations. To address this, we propose ThinkGuard, a critique-augmented guardrail model that distills knowledge from high-capacity LLMs by generating structured critiques alongside safety labels. Fine-tuned on critique-augmented data, the captured deliberative thinking ability drastically enhances the guardrail's cautiousness and interpretability. Evaluated on multiple safety benchmarks, ThinkGuard achieves the highest average F1 and AUPRC, outperforming all baselines. Compared to LLaMA Guard 3, ThinkGuard improves accuracy by 16.1% and macro F1 by 27.0%. Moreover, it surpasses label-only fine-tuned models, confirming that structured critiques enhance both classification precision and nuanced safety reasoning while maintaining computational efficiency.

Key facts. Cited by 29 papers in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 24 of its 55 references that are in the map. First posted on arXiv 19 Feb 2025.

Builds on

24 of 55 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023DUMA: a Dual-Mind Conversational Agent with Fast and Slow Thinking27 Oct

Other references · 23

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  4. 2020Scaling Laws for Neural Language Modelscited by 6,411
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  6. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  7. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,124
  8. 2024DeepSeek-V3 Technical Reportcited by 724
  9. 2022Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learnedcited by 702
  10. 2023Certifying LLM Safety against Adversarial Promptingcited by 248
  11. 2024ShieldGemma: Generative AI Content Moderation Based on Gemmacited by 224
  12. 2024Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-thinking Reasoning Systemscited by 162
11 more
  1. 2022Thinking Fast and Slow in Large Language Modelscited by 136
  2. 2024AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Expertscited by 126
  3. 2024Llama Guard 3 Vision: Safeguarding Human-AI Image Understanding Conversationscited by 118
  4. 2024Self-Generated Critiques Boost Reward Modeling for Language Modelscited by 69
  5. 2024Unveiling the Safety of GPT-4o: An Empirical Study using Jailbreak Attackscited by 35
  6. 2024Securing Large Language Models: Addressing Bias, Misinformation, and Prompt Attackscited by 24
  7. 2024Llama Guard 3-1B-INT4: Compact and Efficient Safeguard for Human-AI Conversationscited by 24
  8. 2024Conceptual and Unbiased Reasoning in Language Modelscited by 19
  9. 2022LAMBRETTA: Learning to Rank for Twitter Soft Moderationcited by 8
  10. 2024Merging Improves Self-Critique Against Jailbreak Attackscited by 8
  11. 2024LLM Safeguard is a Double-Edged Sword: Exploiting False Positives for Denial-of-Service Attackscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 15; 2026: 14.

29 papers.

Sort:

By area:

  1. 2025From System 1 to System 2: A Survey of Reasoning Large Language Modelscited by 319
  2. 2025Safety in Large Reasoning Models: A Surveycited by 58
  3. 2025Contextual Integrity in LLMs via Reasoning and Reinforcement Learningcited by 30
  4. 2025100 Days After DeepSeek-R1: A Survey on Replication Studies and More Directions for Reasoning Language Modelscited by 29
  5. 2025X-Guard: Multilingual Guard Agent for Content Moderationcited by 13
  6. 2025RedCoder: Automated Multi-Turn Red Teaming for Code LLMscited by 8
  7. 2025Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Modelscited by 7
  8. 2025UnsafeChain: Enhancing Reasoning Model Safety via Hard Casescited by 7
  9. 2025Deep Research with Open-Domain Evaluation and Multi-Stage Guardrails for Safetycited by 6
  10. 2026Structured Security Auditing and Robustness Enhancement for Untrusted Agent Skillscited by 6
  11. 2025ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMscited by 5
  12. 2025A Survey on Collaborating Small and Large Language Models for Performance, Cost-effectiveness, Cloud-edge Privacy, and Trustworthinesscited by 5
17 more
  1. 2026LPG: Balancing Efficiency and Policy Reasoning in Latent Policy Guardrailscited by 4
  2. 2025Attention-Aware GNN-based Input Defense against Multi-Turn LLM Jailbreakcited by 2
  3. 2025Learning Efficient Guardrails for Compliancecited by 2
  4. 2026Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representationscited by 2
  5. 2025From Perception to Reasoning: Deep Thinking Empowers Multimodal Large Language Modelscited by 1
  6. 2026Spider-Sense: Intrinsic Risk Sensing for Efficient Agent Defense with Hierarchical Adaptive Screeningcited by 1
  7. 2026ModelLens: Finding the Best for Your Task from Myriads of Modelscited by 1
  8. 2026Triaging Threats to Specialized Guardrailscited by 1
  9. 2026ConsisGuard: Aligning Safety Deliberation with Policy Enforcement in LLM Guardrailscited by 1
  10. 2026Do Safety Guardrails Need to Reason? LeanGuard: A Fast and Light Approach for Robust Moderationcited by 1
  11. 2025From Rogue to Safe AI: The Role of Explicit Refusals in Aligning LLMs with International Humanitarian Lawcited by 0
  12. 2026SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screeningcited by 0
  13. 2026Exploring and Developing a Pre-Model Safeguard with Draft Modelscited by 0
  14. 2026Robust and Efficient Guardrails with Latent Reasoningcited by 0
  15. 2026Risk-Aware LLM Agents for Geospatial Data Retrieval: Design and Preliminary Adversarial Evaluationcited by 0
  16. 2026Securing Multi-Agent GIS Systems: Risk Evaluation and Prompt Hardening Optimizationcited by 0
  17. 2026PolicyShiftGuard: Benchmarking and Improving Policy-Adaptive Image Guardrailscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026YuFeng-XGuard: A Reasoning-Centric, Interpretable, and Flexible Guardrail Model for Large Language Models
  2. 2026DT-Guard: Intent-Driven Reasoning-Active Training for Reasoning-Free LLM Safety Guardrail
  3. 2025Safety Through Reasoning: An Empirical Study of Reasoning Guardrail Models
  4. 2026TRACE: An Evidence-Grounded Benchmark for Safety Evaluation of Large Reasoning Models
  5. 2026CourtGuard: A Model-Agnostic Framework for Zero-Shot Policy Adaptation in LLM Safety
  6. 2026Safety Reconstructed: Generative Modeling via Masked Diffusion Builds Strong Safety Guardrails
  7. 2026SentGuard: Sentence-Level Streaming Guardrails for Large Language Models
  8. 2025SafeRBench: Dissecting the Reasoning Safety of Large Language Models