# RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models

arXiv: https://arxiv.org/abs/2510.10390 · HTML: https://mlsanity.com/paper/2510.10390/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Hallucination detection in RAG, one of 15 directions in Retrieval-augmented generation.
- Builds on 26 of its 47 references that are in the map.
- First posted on arXiv 12 Oct 2025.

## Placed in

- Retrieval-augmented generation / [Hallucination detection in RAG](https://mlsanity.com/rag/hallucination-detection-in-rag/)

## Builds on (most cited first, 26 of 26)

- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models](https://mlsanity.com/paper/2206.04615/) (2022)
- [A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions](https://mlsanity.com/paper/2311.05232/) (2023)
- [Know What You Don't Know: Unanswerable Questions for SQuAD](https://mlsanity.com/paper/1806.03822/) (2018)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Teaching Models to Express Their Uncertainty in Words](https://mlsanity.com/paper/2205.14334/) (2022)
- [BBQ: A Hand-Built Bias Benchmark for Question Answering](https://mlsanity.com/paper/2110.08193/) (2021)
- [KILT: a Benchmark for Knowledge Intensive Language Tasks](https://mlsanity.com/paper/2009.02252/) (2020)
- [A Dataset of Information-Seeking Questions and Answers Anchored in Research Papers](https://mlsanity.com/paper/2105.03011/) (2021)
- [AmbigQA: Answering Ambiguous Open-domain Questions](https://mlsanity.com/paper/2004.10645/) (2020)
- [Hallucination is Inevitable: An Innate Limitation of Large Language Models](https://mlsanity.com/paper/2401.11817/) (2024)
- [FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation](https://mlsanity.com/paper/2310.03214/) (2023)
- [MediQ: Question-Asking LLMs and a Benchmark for Reliable Interactive Clinical Reasoning](https://mlsanity.com/paper/2406.00922/) (2024)
- [SituatedQA: Incorporating Extra-Linguistic Contexts into QA](https://mlsanity.com/paper/2109.06157/) (2021)
- [AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions](https://mlsanity.com/paper/2506.09038/) (2025)
- [Know Your Limits: A Survey of Abstention in Large Language Models](https://mlsanity.com/paper/2407.18418/) (2024)
- [The Art of Saying No: Contextual Noncompliance in Language Models](https://mlsanity.com/paper/2407.12043/) (2024)
- [FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows"](https://mlsanity.com/paper/2410.03727/) (2024)
- [Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models](https://mlsanity.com/paper/2305.13712/) (2023)
- [(QA)2: Question Answering with Questionable Assumptions](https://mlsanity.com/paper/2212.10003/) (2022)
- [Won't Get Fooled Again: Answering Questions with False Premises](https://mlsanity.com/paper/2307.02394/) (2023)
- [ALCUNA: Large Language Models Meet New Knowledge](https://mlsanity.com/paper/2310.14820/) (2023)
- [WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models](https://mlsanity.com/paper/2311.15930/) (2023)
- [GaRAGe: A Benchmark with Grounding Annotations for RAG Evaluation](https://mlsanity.com/paper/2506.07671/) (2025)
- [ELOQ: Resources for Enhancing LLM Detection of Out-of-Scope Questions](https://mlsanity.com/paper/2410.14567/) (2024)

## Built on by (most cited first, 9 of 9)

- [HopRefusalBench: Diagnosing Refusal Failures in Search-Augmented Agents for Multi-Hop Reasoning](https://mlsanity.com/paper/2608.01358/) (2026)
- [Knowledge Distillation Must Account for What It Loses](https://mlsanity.com/paper/2604.25110/) (2026)
- [Prompt-Based Abstention Fails Under Misleading Context: A Controlled Study of Small Frozen RAG Models](https://mlsanity.com/paper/2608.22228/) (2026)
- [Bounding Hallucinations: Merlin-Arthur Protocols for Mutual-Information Bounds in Language Models](https://mlsanity.com/paper/2512.11614/) (2025)
- [Don't Start What You Can't Finish: A Counterfactual Audit of Support-State Triage in LLM Agents](https://mlsanity.com/paper/2604.16752/) (2026)
- [MMOOC: A Comprehensive Benchmark for Out-of-Context Evaluation in Multimodal Large Language Models](https://mlsanity.com/paper/2607.27637/) (2026)
- [How AI Assistants Respond to Repeated Abuse](https://mlsanity.com/paper/2609.17547/) (2026)
- [The Commit-Abstain Circuit: Why Language Models Hallucinate Instead of Abstaining](https://mlsanity.com/paper/2609.32964/) (2026)
- [CoverSynth: Coverage-Guided Generation of Retrieval Evaluation Sets](https://mlsanity.com/paper/2610.15409/) (2026)

## Abstract (arXiv)

The ability of language models in RAG systems to selectively refuse to answer based on flawed context is critical for safety, yet remains a significant failure point. Our large-scale study reveals that even frontier models struggle in this setting, with refusal accuracy dropping below 50% on multi-document tasks, while exhibiting either dangerous overconfidence or overcaution. Static benchmarks fail to reliably evaluate this capability, as models exploit dataset-specific artifacts and memorize test instances. We introduce RefusalBench, a generative methodology that programmatically creates diagnostic test cases through controlled linguistic perturbation. Our framework employs 176 distinct perturbation strategies across six categories of informational uncertainty and three intensity levels. Evaluation of over 30 models uncovers systematic failure patterns: refusal comprises separable detection and categorization skills, and neither scale nor extended reasoning improves performance. We find that selective refusal is a trainable, alignment-sensitive capability, offering a clear path for improvement. We release two benchmarks -- RefusalBench-NQ (single document) and RefusalBench-GaRAGe (multi-document) -- and our complete generation framework to enable continued, dynamic evaluation of this critical capability.