---
title: "Discussion on \"RAG poisoning: how one document can hijack your AI agent\" | Hashnode"
description: "Discussion on \"RAG poisoning: how one document can hijack your AI agent\". Indirect prompt injection is the attack your prompt filter never sees. Here's how it works — with a concrete example — and how to defend against it."
url: "https://hashnode.com/posts/rag-poisoning-how-one-document-can-hijack-your-ai-agent/6a3854da482fd51aa6af25ef"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [RAG poisoning: how one document can hijack your AI agent](https://trustgateai.hashnode.dev/rag-poisoning-how-one-document-can-hijack-your-ai-agent?utm_source=hashnode&utm_medium=feed)

Indirect prompt injection is the attack your prompt filter never sees. Here's how it works — with a concrete example — and how to defend against it.

Jun 21 · 5 min read · trustgateai.hashnode.dev  
Read the article

-   [#ai-llm-rag-machine-learning-vector-databases-nlp](https://hashnode.com/tag/ai-llm-rag-machine-learning-vector-databases-nlp)
-   [#security](https://hashnode.com/tag/security)
-   [#rag](https://hashnode.com/tag/rag)
-   [#ai-agent](https://hashnode.com/tag/ai-agent)
-   [#ai-security](https://hashnode.com/tag/ai-security)

## Written by

[Trustgate AI](https://hashnode.com/@trustgateai)

@trustgateai · Cary, NC

  
[View profile](https://hashnode.com/@trustgateai)

Self-hosted AI security gateway for agentic AI. Inspect every surface — prompt, RAG, tools, MCP, egress — with zero data egress.

The self-hosted trust plane for agentic AI. Secure every surface of your AI agents — see none of your data.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
