Stance Poisoning and Evidence Balancing in RAG Systems

# Stance Poisoning and Evidence Balancing in RAG Systems

https://gcdradar.vercel.app/incidents/stance-poisoning-and-evidence-balancing-in-rag-systems · AIR-2026-1564 · closed 2026-11-10

- Lane: Research
- Severity: 1 (Research / proof of concept): Research paper demonstrating a novel stance poisoning attack on RAG models.
- Fix status: Fix status unknown
- First reported: 2026-10-27
- AI role: Attack on AI
- Products: Retrieval-Augmented Generation Systems

## Summary

Researchers demonstrate that injecting correct facts with a one-sided stance into retrieval sources can systematically alter the output stance of retrieval-augmented generation models. Tests across multiple vendor models revealed consistent stance shifts.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-10-27: Factual Yet One-Sided: Stance Poisoning and Evidence Balancing in Retrieval-Augmented Generation (https://arxiv.org/abs/2610.34294)

## Sources

- 2026-10-27 arXiv: Factual Yet One-Sided: Stance Poisoning and Evidence Balancing in Retrieval-Augmented Generation https://arxiv.org/abs/2610.34294

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.