Biomedical KG-RAG Poisoning Dataset: Corpus and Poison Files for Blast Radius Evaluation in GraphRAG Systems
Abstract
The data presented in this study were generated to support the empirical evaluation of triple-level poisoning attacks in Knowledge Graph Retrieval-Augmented Generation (KG-RAG) systems, with a focus on quantifying blast radius, defined as the number of distinct query responses corrupted per injected false triple. No publicly available dataset exists for evaluating cascade corruption in graph-based RAG pipelines, where a single poisoned relational triple can propagate through multiple inference paths simultaneously. The dataset was created as part of a broader study on structural attack surfaces in GraphRAG systems and is designed to enable controlled, reproducible experimentation on cross-query cascade effects under three structurally motivated attack scenarios. The dataset consists of 13 plain-text files organized into two categories: ten biomedical corpus files and three poison injection files. The corpus files were retrieved from PubMed using the Biopython Entrez API, with five abstracts per biomedical topic cluster across ten search queries, yielding approximately 50 documents in total. The three poison files contain synthetically authored false biomedical assertions corresponding to edge poisoning, hub poisoning, and chain poisoning attack scenarios respectively. The dataset supports evaluation of structural poisoning attacks and cascade corruption measurement in knowledge graph retrieval systems.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.