Entity · paper

Deceptive Grounding: Entity Attribution Failure in Clinical Retrieval-Augmented Generation

paperactiveprovisionaldeceptive-grounding-entity-attribution-failure-in-clinical-retrieval-augmented-generation-db340caf·1 events·first seen 10h ago

Aliases: Deceptive Grounding: Entity Attribution Failure in Clinical Retrieval-Augmented Generation

Co-occurring entities

deceptive grounding

More like this (12)

Retrieval-Augmented Generation Tracing Target Answers in Poisoned Retrieval Corpora via Token Influence Attribution referential grounding Achieving Precise Text-To-Cypher Via Grounded Knowledge Graph Data Generation Compositional Reasoning Depth Predicts Clinical AI Failure Multimodal Augmented Generation via Multimodal Retrieval Workshop Knowledge-Graph Grounding Helps LLMs Only for Out-of-Training Knowledge: A Controlled Study on Clinical Question Answering Text Generation Inference Recalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented Models A Computational Audit of Demographic Association Encoding in ClinicalBERT Language Predictions generative language modeling From Black-Box to Clinical Insight: A Multi-Stage Explainable Framework for Speech-Based Cognitive Impairment Detection

Recent events (1)

7arXiv · cs.CL·10h ago·source ↗

Deceptive Grounding: Entity Attribution Failure in Clinical RAG Systems Evades Standard Checks

A new arXiv paper identifies 'deceptive grounding' (DG), a failure mode in clinical retrieval-augmented generation where a model presents evidence about drug Y as evidence about queried drug X, passing all standard faithfulness, hallucination, and citation checks. Using a controlled factorial benchmark across 13 models, the authors find DG rates of 8–87% under adversarial conditions, with medical fine-tuned models reaching 86.7%—worse than general models. A production measurement across 740 drug-disease pairs finds 7.8% overall DG in a deployed system, rising to 13.6% for recently approved drugs. The paper proposes entity-attribution verification as a mitigation, achieving 97.0% precision and 98.7% recall, and notes no existing RAG evaluation framework implements this check.

Evaluation and Benchmarking AI Safety Research deceptive grounding Deceptive Grounding: Entity Attribution Failure in Clinical Retrieval-Augmented Generation +1 more