When Poison Fails After Retrieval: Revisiting Corpus Poisoning under Chunking and Reranking Pipelines
arXiv:2606.11265v1 Announce Type: cross Abstract: Retrieval-Augmented Generation (RAG) systems are vulnerable to corpus poisoning attacks that manipulate downstream model outputs through malicious knowledge injection. Existing studies mainly evaluate poisoning under simplified retrieval settings, overlooking practical RAG pipelines involving document chunking, dense retrieval, reranking, and grounded generation. In this paper, we revisit corpus poisoning under realistic multi-stage retrieval pipelines and show that many existing attacks substantially degrade after reranking despite achieving h
This research is emerging now as RAG systems are becoming more prevalent and sophisticated, exposing new vectors for adversarial attacks that need to be understood for robust deployment.
Sophisticated readers should care because this research highlights critical vulnerabilities in AI systems, especially Retrieval-Augmented Generation (RAG), which can be exploited to manipulate outputs and undermine trust.
The understanding of RAG system vulnerabilities changes as the efficacy of existing corpus poisoning attacks is shown to vary significantly under realistic multi-stage retrieval pipelines.
- · AI security researchers
- · Developers of robust RAG systems
- · Organizations prioritizing AI safety and integrity
- · Malicious actors relying on simplified poisoning techniques
- · Organizations with unhardened RAG deployments
- · Developers neglecting advanced adversarial testing
Existing corpus poisoning attacks may be less effective against highly engineered RAG pipelines.
This will drive the development of more advanced and adaptive poisoning techniques, or more robust RAG defenses.
Increased focus on end-to-end security for AI systems will likely emerge, incorporating a 'defense-in-depth' approach.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI