Abstract
While retrieval-augmented generation systems partially address the hallucination issues in large language models, it also introduces new vulnerabilities to knowledge corruption attacks. Adversaries exploit these vulnerabilities by poisoning documents provided by RAG system to manipulate LLM outputs. To counter this threat, we propose SecureCollaRAG, a Byzantine-tolerant collaborative RAG framework leveraging Multi-source Knowledge Validation Mechanism. Our approach enables agent system to securely verify document provenance through dynamic GNN-based credibility scoring, effectively preventing stealthy knowledge corruption attacks while preserving essential domain knowledge integrity. Through extensive evaluations and formal analysis, we demonstrate that SecureCollaRAG maintains robustness against attackers under non-IID data distributions. © 2026 Owner/Author.
| Original language | English |
|---|---|
| Title of host publication | WWW '26 - Proceedings of the ACM Web Conference 2026 |
| Place of Publication | New York, NY |
| Publisher | Association for Computing Machinery |
| Pages | 2661-2672 |
| Number of pages | 12 |
| ISBN (Print) | 9798400723070 |
| DOIs | |
| Publication status | Published - Apr 2026 |
| Event | 35th ACM Web Conference (WWW 2026) - Dubai, United Arab Emirates Duration: 29 Jun 2026 → 3 Jul 2026 |
Publication series
| Name | WWW - Proceedings of the ACM Web Conference |
|---|
Conference
| Conference | 35th ACM Web Conference (WWW 2026) |
|---|---|
| Abbreviated title | WWW ’26 |
| Place | United Arab Emirates |
| City | Dubai |
| Period | 29/06/26 → 3/07/26 |
Funding
This paper is supported by National Key Research and Development Program of China under the grant No.2023YFF0905300, National Natural Science Foundation of China under Grant U2468205.
Research Keywords
- graph neural network
- knowledge corruption attacks
- large language model
- retrieval-augmented generation
- securecollarag
Publisher's Copyright Statement
- This full text is made available under CC-BY 4.0. https://creativecommons.org/licenses/by/4.0/
Fingerprint
Dive into the research topics of 'Combating Knowledge Corruption in Agent Systems: A Byzantine-Tolerant Secure Collaborative RAG Framework'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver