Day 8 of 1560 minExploit ArchitectHands-on lab
Day 8: Indirect Injection Through Documents and Retrieval
Taught by Mr Elite Updated
By the end of today you can
- Explain the difference between direct and indirect injection by naming attacker, victim and ingestion path
- Craft a document with hidden or specially formatted text that a summariser obeys, using a harmless canary
- Plant a poisoned article in a RAG knowledge base and show it is retrieved for another user
- Recommend detections and fixes for ingestion-path injection
Picture a support team that lets staff upload supplier PDFs into an AI summariser. Nobody types anything hostile into the chat box. The attacker never touches the chat box at all. They wrote one line in a document weeks ago, and today a colleague asks for a summary and the model quietly follows the line instead.
That is indirect injection. The person typing is the victim, not the attacker. Today you practise it in two sandboxed SecurityElites labs, using harmless canary strings, and then you learn how a defender spots and stops it.
In this lesson
- Direct versus indirect: who is who
- Documents: hidden text and trusted structures
- Watch me do it: a canary in a Markdown file
- RAG poisoning and ranking
- Detect and fix
Premium course
$27 one-time, lifetime access
This day is part of the full course. Unlock it to read every lesson, do every mission and track your progress to the end.