Adversarial Threat Vectors and Risk Mitigation for Retrieval-Augmented Generation Systems

TL;DR

This paper identifies key adversarial threats to Retrieval-Augmented Generation systems, analyzes their risks, and proposes mitigation strategies like input validation and real-time monitoring to enhance system security.

Contribution

It provides a comprehensive analysis of adversarial attack vectors on RAG systems and introduces a prioritized control list for risk mitigation.

Findings

Prompt injection is a major threat.

Adversarial training improves robustness.

Real-time monitoring helps detect attacks.

Abstract

Retrieval-Augmented Generation (RAG) systems, which integrate Large Language Models (LLMs) with external knowledge sources, are vulnerable to a range of adversarial attack vectors. This paper examines the importance of RAG systems through recent industry adoption trends and identifies the prominent attack vectors for RAG: prompt injection, data poisoning, and adversarial query manipulation. We analyze these threats under risk management lens, and propose robust prioritized control list that includes risk-mitigating actions like input validation, adversarial training, and real-time monitoring.