Knowing When Not to Answer: Lightweight KB-Aligned OOD Detection for Safe RAG

ACL ARR 2026 January Submission4775 Authors

Keywords: dialogue systems, safety, retrieval augmented generation, out of domain detection

Abstract:

Retrieval-Augmented Generation (RAG) systems are increasingly deployed in high-stakes domains, where safety depends not only on \emph{how} a system answers, but also on \emph{whether} a query should be answered given a knowledge base (KB). Out-of-domain (OOD) queries can cause dense retrieval to surface weakly related context and lead the generator to produce fluent but unjustified responses. We study lightweight, KB-aligned OOD detection as an always-on gate for RAG systems. Our approach applies PCA to KB embeddings and scores queries in a compact subspace selected either by explained-variance retention (EVR) or by a separability-driven $t$-test ranking. We evaluate geometric semantic-search rules and lightweight classifiers across 16 domains, including high-stakes COVID-19 and Substance Use KBs, and stress-test robustness using both LLM-generated attacks and an in-the-wild 4chan attack.
We find that low-dimensional detectors achieve competitive OOD performance while being faster, cheaper, and more interpretable than prompted LLM-based judges. Finally, human and LLM-based evaluations show that OOD queries primarily degrade the \emph{relevance} of RAG outputs, showing the need for efficient external OOD detection to maintain safe, in-scope behavior.

Paper Type: Long

Research Area: Dialogue and Interactive Systems

Research Area Keywords: dialogue systems, safety, retrieval augmented generation, out of domain detection

Contribution Types: Model analysis & interpretability, Data analysis

Languages Studied: English

Submission Number: 4775