RAG Shield: A Multi-Layer Defense System Against Poisoning Attacks in Retrieval-Augmented Generation

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Petti, Fabio
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866902202541408256
author Petti, Fabio
author_facet Petti, Fabio
contents <p>This whitepaper presents RAG Shield, a security-focused framework for<br>defending Retrieval-Augmented Generation (RAG) pipelines against<br>poisoning and adversarial manipulation at the retrieval layer.</p> <p>The work introduces a multi-layer defense architecture combining<br>cryptographic document provenance validation, semantic anomaly detection,<br>and secure, authority-weighted retrieval control. A realistic threat<br>model is defined, focusing on poisoning of retrieval corpora rather than<br>prompt or model-level attacks. The system is evaluated against multiple<br>attack scenarios under controlled conditions.</p> <p>RAG Shield is designed as a framework-agnostic security control layer<br>that operates independently of the underlying language model and vector<br>database, enabling deployment in enterprise and regulated environments<br>without modification of existing RAG architectures.</p> <p>This document is released as a technical preprint to establish prior art<br>and support open discussion in the areas of AI security, adversarial<br>machine learning, and secure enterprise RAG deployment.</p> <p>Project website and system overview:<br>https://sentinelrag.com</p> <p>Contact:<br>info@sentinelrag.com</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18449664
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle RAG Shield: A Multi-Layer Defense System Against Poisoning Attacks in Retrieval-Augmented Generation
Petti, Fabio
RAG security
retrieval-augmented generation
adversarial machine learning
poisoning attacks
AI security
<p>This whitepaper presents RAG Shield, a security-focused framework for<br>defending Retrieval-Augmented Generation (RAG) pipelines against<br>poisoning and adversarial manipulation at the retrieval layer.</p> <p>The work introduces a multi-layer defense architecture combining<br>cryptographic document provenance validation, semantic anomaly detection,<br>and secure, authority-weighted retrieval control. A realistic threat<br>model is defined, focusing on poisoning of retrieval corpora rather than<br>prompt or model-level attacks. The system is evaluated against multiple<br>attack scenarios under controlled conditions.</p> <p>RAG Shield is designed as a framework-agnostic security control layer<br>that operates independently of the underlying language model and vector<br>database, enabling deployment in enterprise and regulated environments<br>without modification of existing RAG architectures.</p> <p>This document is released as a technical preprint to establish prior art<br>and support open discussion in the areas of AI security, adversarial<br>machine learning, and secure enterprise RAG deployment.</p> <p>Project website and system overview:<br>https://sentinelrag.com</p> <p>Contact:<br>info@sentinelrag.com</p>
title RAG Shield: A Multi-Layer Defense System Against Poisoning Attacks in Retrieval-Augmented Generation
topic RAG security
retrieval-augmented generation
adversarial machine learning
poisoning attacks
AI security
url https://doi.org/10.5281/zenodo.18449664