DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair
On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks
also in 1 more
Abstract (arXiv)
While safety alignment and guardrails help large language models (LLMs) avoid harmful outputs, they can also induce overrefusal, i.e., unwarranted rejection of benign queries that merely appear risky. We present DDOR (Delta Debugging for OverRefusal), a fully automated and explainable framework for overrefusal testing and repair in a black-box setting, where only model inputs and outputs are accessible and internal safety mechanisms remain opaque. DDOR applies delta debugging to localize minimal refusal-triggering fragments (mRTFs) that provide phrase-level, explainable evidence for why a refusal occurs. Conditioned on these mRTFs, DDOR generates diverse, context-rich prompts and performs multi-oracle validation to filter intrinsically unsafe or ambiguous cases, producing scalable and model-specific overrefusal test suites (approximately 1K cases per model). Beyond evaluation, we further leverage localized mRTFs to perform targeted prompt repair, substantially reducing overrefusal while preserving the original intent and maintaining safety on genuinely harmful inputs. Overall, DDOR offers a practical end-to-end solution to both evaluate and mitigate overrefusal, improving LLM usability without sacrificing safety.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 14 of its 31 references that are in the map. First posted on arXiv 2 Jun 2026.
Builds on
14 of 31 references.
Closest ancestors · 2
Same direction, newest first
- 2025Improving LLM Safety Alignment with Dual-Objective Optimization5 Mar
- 2024SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models18 Jun
Other references · 12
- 2022Training language models to follow instructions with human feedbackcited by 15,971
- 2022Constitutional AI: Harmlessness from AI Feedbackcited by 3,050
- 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
- 2023SafetyBench: Evaluating the Safety of Large Language Modelscited by 199
- 2025GuardReasoner: Towards Reasoning-based LLM Safeguardscited by 100
- 2024Steering Language Model Refusal with Sparse Autoencoderscited by 95
- 2024Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablationcited by 48
- 2024S-Eval: Towards Automated and Comprehensive Safety Evaluation for Large Language Modelscited by 33
- 2020Importance-Driven Deep Learning System Testingcited by 24
- 2025Understanding and Mitigating Overrefusal in LLMs from an Unveiling Perspective of Safety Decision Boundarycited by 18
- 2025Advancing LLM Safe Alignment with Safety Representation Rankingcited by 16
- 2025Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behaviorcited by 12
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025Understanding and Mitigating Over-refusal for Large Language Models via Representation Intervention
- 2025ORFuzz: Fuzzing the "Other Side" of LLM Safety -- Testing Over-Refusal
- 2026Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment
- 2026SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs
- 2025FalseReject: A Resource for Improving Contextual Safety and Mitigating Over-Refusals in LLMs via Structured Reasoning
- 2026Beyond Token Positions: Safety Alignment Across Denoising Steps in Diffusion Language Models
- 2024Jailbreak Antidote: Runtime Safety-Utility Balance via Sparse Representation Adjustment in Large Language Models
- 2026RAS: Measuring LLM Safety Through Refusal Alignment