# Prompt-Based Abstention Fails Under Misleading Context: A Controlled Study of Small Frozen RAG Models

arXiv: https://arxiv.org/abs/2608.22228 · HTML: https://mlsanity.com/paper/2608.22228/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in RAG robustness and adaptive retrieval, one of 15 directions in Retrieval-augmented generation.
- Builds on 26 of its 39 references that are in the map.
- First posted on arXiv 23 Aug 2026.

## Placed in

- Retrieval-augmented generation / [RAG robustness and adaptive retrieval](https://mlsanity.com/rag/rag-robustness-and-adaptive-retrieval/)

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

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://mlsanity.com/paper/1809.09600/) (2018)
- [Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection](https://mlsanity.com/paper/2310.11511/) (2023)
- [DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing](https://mlsanity.com/paper/2111.09543/) (2021)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://mlsanity.com/paper/2302.00093/) (2023)
- [Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback](https://mlsanity.com/paper/2305.14975/) (2023)
- [C-Pack: Packed Resources For General Chinese Embeddings](https://mlsanity.com/paper/2309.07597/) (2023)
- [Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs](https://mlsanity.com/paper/2503.01743/) (2025)
- [Entity-Based Knowledge Conflicts in Question Answering](https://mlsanity.com/paper/2109.05052/) (2021)
- [Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts](https://mlsanity.com/paper/2305.13300/) (2023)
- [Don't Hallucinate, Abstain: Identifying LLM Knowledge Gaps via Multi-LLM Collaboration](https://mlsanity.com/paper/2402.00367/) (2024)
- [What Evidence Do Language Models Find Convincing?](https://mlsanity.com/paper/2402.11782/) (2024)
- [Sufficient Context: A New Lens on Retrieval Augmented Generation Systems](https://mlsanity.com/paper/2411.06037/) (2024)
- [Retrieval-Augmented Generation with Conflicting Evidence](https://mlsanity.com/paper/2504.13079/) (2025)
- [Knowing What LLMs DO NOT Know: A Simple Yet Effective Self-Detection Method](https://mlsanity.com/paper/2310.17918/) (2023)
- [DYNAMICQA: Tracing Internal Knowledge Conflicts in Language Models](https://mlsanity.com/paper/2407.17023/) (2024)
- [Characterizing LLM Abstention Behavior in Science QA with Context Perturbations](https://mlsanity.com/paper/2404.12452/) (2024)
- [Open Domain Question Answering with Conflicting Contexts](https://mlsanity.com/paper/2410.12311/) (2024)
- [A Novel Characterization of the Population Area Under the Risk Coverage Curve (AURC) and Rates of Finite Sample Estimators](https://mlsanity.com/paper/2410.15361/) (2024)
- [Towards More Robust Retrieval-Augmented Generation: Evaluating RAG Under Adversarial Poisoning Attacks](https://mlsanity.com/paper/2412.16708/) (2024)
- [Why Uncertainty Estimation Methods Fall Short in RAG: An Axiomatic Analysis](https://mlsanity.com/paper/2505.07459/) (2025)
- [Controlling Risk of Retrieval-augmented Generation: A Counterfactual Prompting Framework](https://mlsanity.com/paper/2409.16146/) (2024)
- [RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models](https://mlsanity.com/paper/2510.10390/) (2025)

## Built on by (most cited first, 1 of 1)

- [Relevance Is Not Sufficient Evidence: Detecting Evidence Gaps Before Generation in RAG](https://mlsanity.com/paper/2609.37469/) (2026)

## Abstract (arXiv)

Missing and misleading evidence are not the same problem in retrieval-augmented generation (RAG), but prompt-based abstention treats them alike. Models abstain when context is absent, not when it is misleading. We introduce GRAB-RAG (Graded Abstention Benchmark for Retrieval-Augmented Generation), a paired benchmark that tests the same questions across four context conditions (supportive, degraded, missing, and misleading) in Natural Questions and HotpotQA. In the misleading condition, we edit a gold passage to support a wrong answer and place it among other retrieved passages. We test five abstention policies on three small frozen models (3.8B--8B) across two QA benchmarks. Models abstain reliably when evidence is missing, but under explicit abstention prompting still answer 41.6% of misleading questions, with 63% of those answers echoing the planted wrong entity verbatim. Chain-of-thought provides little additional benefit. A generator-side conflict check cuts the rate to 13.3% but discards many correct answers, while an NLI verifier recovers that coverage but fails when parametric memory and the misleading passage agree on the same wrong answer. Prompt-based abstention asks whether context is sufficient, not whether it is correct. Neither verifier closes this gap without trading coverage for safety.