← Concept IndexDEFINITION WHY IT MATTERS COMMONLY CONFUSED WITH SOURCES
Indirect prompt injection
Also called: poisoned content, injection via documents
Injection that arrives not from the user but from content the agent fetches on its own — a booby-trapped web page, a shared doc, a calendar invite — often with the malicious text hidden from human eyes.
This is the version that scales: an attacker never talks to you, they just plant instructions where your agent will read them. It's the reason 'the agent browsed a page' is a security event, not a convenience.
Direct prompt injection, where the user themselves types the malicious instruction. Here the user is the victim, not the source.
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1U.S. National Institute of Standards and Technology · 2023-01-26
First-pass citations, limited to primary sources; a reviewer will broaden and verify these before this entry leaves draft.