mlsanityData to 14 Oct 2026

Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behavior

arXiv 2503.17882 ↗First posted 22 Mar 2025cs.CL cs.AIcited by 12

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

Abstract (arXiv)

Recent advancements in large language models (LLMs) have demonstrated that fine-tuning and human alignment can render LLMs harmless. In practice, such "harmlessness" behavior is mainly achieved by training models to reject harmful requests, such as "Explain how to burn down my neighbor's house", where the model appropriately declines to respond. However, this approach can inadvertently result in false refusal, where models reject benign queries as well, such as "Tell me how to kill a Python process". In this work, we demonstrate that prompting safety reflection before generating a response can mitigate false refusal behavior. Building on this finding, we introduce the Think-Before-Refusal (TBR) schema and conduct safety-aware instruction fine-tuning incorporating safety reflection. In an ablation study across 15 pre-trained models, we show that models fine-tuned with safety reflection significantly reduce false refusal behavior while maintaining safety and overall performance compared to those fine-tuned without safety reflection.

Key facts. Cited by 12 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 25 of its 25 references that are in the map. First posted on arXiv 22 Mar 2025.

Builds on

25 of 25 references.

Closest ancestors · 2

Same direction, newest first

  1. 2024Deliberative Alignment: Reasoning Enables Safer Language Models20 Dec
  2. 2023Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinking16 Nov

Other references · 23

  1. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  5. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  6. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  7. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  8. 2018CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledgecited by 1,822
  9. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  10. 2023LIMA: Less Is More for Alignmentcited by 1,100
  11. 2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 656
  12. 2023The Falcon Series of Open Language Modelscited by 535
11 more
  1. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 451
  2. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 387
  3. 2019Explain Yourself! Leveraging Language Models for Commonsense Reasoningcited by 367
  4. 2023An Overview of Catastrophic AI Riskscited by 222
  5. 2022On Second Thought, Let's Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoningcited by 194
  6. 2020Recipes for Safety in Open-domain Chatbotscited by 170
  7. 2023Self-Guard: Empower the LLM to Safeguard Itselfcited by 54
  8. 2024SCANS: Mitigating the Exaggerated Safety for LLMs via Safety-Conscious Activation Steeringcited by 50
  9. 2024Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablationcited by 48
  10. 2021Think Before You Speak: Explicitly Generating Implicit Commonsense Knowledge for Response Generationcited by 26
  11. 2024AQA-Bench: An Interactive Benchmark for Evaluating LLMs' Sequential Reasoning Abilitycited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 6; 2026: 6.

12 papers.

  1. 2025Reasoning as an Adaptive Defense for Safetycited by 23
  2. 2025CARE: Decoding Time Safety Alignment via Rollback and Introspection Interventioncited by 6
  3. 2025Read the Scene, Not the Script: Outcome-Aware Safety for LLMscited by 6
  4. 2025Reasoning Up the Instruction Ladder for Controllable Language Modelscited by 5
  5. 2025Beyond Over-Refusal: Scenario-Based Diagnostics and Post-Hoc Mitigation for Exaggerated Refusals in LLMscited by 3
  6. 2025Dynamic Evaluation for Oversensitivity in LLMscited by 2
  7. 2026DDOR: Delta Debugging for Explainable Overrefusal Testing and Repaircited by 0
  8. 2026SHARD: Safe and Helpful Alignment via Self-Reframing Distillationcited by 0
  9. 2026Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Modelscited by 0
  10. 2026Addressing Over-Refusal in LLMs with Competing Rewardscited by 0
  11. 2026Pretraining Curricula Enable Selective Fine-tuningcited by 0
  12. 2026Making Your LLMs More Objective: Stabilizing LLM Safety Behavior Across Traits with Trait-Invariant Safety Tuningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
  2. 2025An Embarrassingly Simple Defense Against LLM Abliteration Attacks
  3. 2026Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models
  4. 2025SafeSwitch: Steering Unsafe LLM Behavior via Internal Activation Signals
  5. 2026LLMs Can Unlearn Refusal with Only 1,000 Benign Samples
  6. 2025Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint
  7. 2026The Refusal--Compliance Tradeoff: A Large-Scale Safety Behavior Audit of Large Language Models
  8. 2026Do Language Models Know When They'll Refuse? Probing Introspective Awareness of Safety Boundaries