# In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning

arXiv: https://arxiv.org/abs/2609.09243 · HTML: https://mlsanity.com/paper/2609.09243/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Privacy and security of RAG, one of 15 directions in Retrieval-augmented generation.
- Builds on 10 of its 10 references that are in the map.
- First posted on arXiv 8 Sep 2026.

## Placed in

- Retrieval-augmented generation / [Privacy and security of RAG](https://mlsanity.com/rag/privacy-and-security-of-rag/)

## Builds on (most cited first, 10 of 10)

- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Billion-scale similarity search with GPUs](https://mlsanity.com/paper/1702.08734/) (2017)
- [REALM: Retrieval-Augmented Language Model Pre-Training](https://mlsanity.com/paper/2002.08909/) (2020)
- [Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection](https://mlsanity.com/paper/2302.12173/) (2023)
- [FEVER: a large-scale dataset for Fact Extraction and VERification](https://mlsanity.com/paper/1803.05355/) (2018)
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://mlsanity.com/paper/2302.00093/) (2023)
- [Poisoning Web-Scale Training Datasets is Practical](https://mlsanity.com/paper/2302.10149/) (2023)
- [PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models](https://mlsanity.com/paper/2402.07867/) (2024)
- [Don't Make Your LLM an Evaluation Benchmark Cheater](https://mlsanity.com/paper/2311.01964/) (2023)

## Abstract (arXiv)

Retrieval-augmented generation (RAG) grounds a language model in retrieved documents, which reduces hallucination but creates a new attack surface: if retrieved text is tampered with, the model may repeat the falsehood. We study how much a small quantized model, Llama 3.1 8B, degrades when a fraction of its retrieved context is poisoned. Three corruption strategies are tested, entity swap, number swap, and negation, each applied to zero, one, two, or three of the three retrieved passages, over a factorial sweep of 588 runs on a fact-checking task built from FEVER. Accuracy falls from 77.9% on clean context to 43.5% when all three passages are corrupted. Entity swap flips the largest share of answers that were correct on clean context. Number-based corruption stays flat while poisoned passages are a minority and jumps once they form a majority, a pattern we re-check with query-level bootstrap intervals. The model rarely invents new falsehoods; its dominant reaction is to abstain, and a lexical overlap proxy of unsupported generation falls under attack rather than rising. The study is a small-scale measurement with coarse automated labels; we treat the strategy contrasts as suggestive until decoding is controlled and stronger adjudication is in place.