Combating Knowledge Corruption in Agent Systems: A Byzantine-Tolerant Secure Collaborative RAG Framework

Zhaoqi Wang, Daqing He, Zijian Zhang, Ye Liu, Jiamou Liu, Zhirui Zeng, Zhan Qin, Zhen Li, Xin Li, Hongwei Yao, Jincheng An, Yong Liu, Yi Li, Qi Sun, Xiulei Liu, and Liehuang Zhu

In Proceedings of the ACM on Web Conference 2026 (WWW), 2026

Abstract: Large language models persistently face the challenge of hallucination suppression. While retrieval-augmented generation systems partially address these issues, 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.

Cite:

@inproceedings{Wang2026CKC,
  author = {Wang, Zhaoqi and He, Daqing and Zhang, Zijian and Liu, Ye and Liu, Jiamou and Zeng, Zhirui and Qin, Zhan and Li, Zhen and Li, Xin and Yao, Hongwei and An, Jincheng and Liu, Yong and Li, Yi and Sun, Qi and Liu, Xiulei and Zhu, Liehuang},
  booktitle = {Proceedings of the ACM on Web Conference 2026 (WWW)},
  month = apr,
  title = {Combating Knowledge Corruption in Agent Systems: A {Byzantine}-Tolerant Secure Collaborative {RAG} Framework},
  year = {2026}
}