keyword

refusal class alignment

Refusal class alignment refers to the property or evaluation criterion in which an artificial intelligence model, upon deciding to decline answering a query due to defective context, attributes its refusal to the specific and correct category of informational flaw. In grounded question answering and retrieval-augmented systems, reference texts may be unanswerable for diverse reasons, such as factual contradictions, insufficient information, temporal invalidity, or linguistic ambiguity. Refusal class alignment evaluates whether a system accurately recognizes and matches the true underlying cause of uncertainty rather than merely deciding to withhold an answer, ensuring that refusals are grounded in a correct understanding of contextual defects rather than arbitrary heuristics, superficial pattern matching, or indiscriminate overcaution.

1 item

RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models

Aashiq Muhamed, Leonardo F. R. Ribeiro, Markus Dreyer, Virginia Smith, Mona T. Diab

OrganizationsAmazonCarnegie Mellon University

Why you should read this

Presents RefusalBench, a generative evaluation framework using 176 perturbation strategies to expose how frontier retrieval-augmented models fail to selectively refuse answering from flawed contexts, while demonstrating that this capability requires targeted training rather than increased model scale.

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.

Source

https://arxiv.org/pdf/2510.10390.pdf

Added

2026-09-30