# Indirect Prompt Injection

> Understand how attackers weaponize external data sources to hijack LLMs without ever directly interacting with the chat interface.

Canonical: https://codelivly.com/rooms/indirect-prompt-injection
Last updated: 2026-10-08

| | |
|---|---|
| Level | Intermediate |
| Category | Cybersecurity |
| Lessons | 4 |
| Estimated time | 30 min |
| XP | 100 |

This module explores indirect prompt injection, a stealthy and highly effective attack vector where malicious instructions are embedded in external data sources (like websites, PDFs, or databases) that the AI later retrieves. You will learn how the "poisoned well" concept works, why Context Authority Bias makes LLMs vulnerable to retrieved data, and how to implement pre-LLM guardrails and strict ingestion controls to secure your RAG pipelines.

## Lessons

1. What is Indirect Prompt Injection and Why It's More Dangerous (30 min)
2. Injecting via Uploaded Documents (30 min)
3. Injecting via Web Content the Model Retrieves (30 min)
4. Injecting via Emails or Third-Party Data Sources (30 min)

## Related

- [Learning path: Prompt Injection Fundamentals](https://codelivly.com/learning-paths/prompt-injection-fundamentals)
- [All rooms](https://codelivly.com/rooms)