indirect

Definition

In indirect prompt injection the attacker never touches the prompt box: they plant instructions in content the model will later retrieve — a web page, a PDF, an email, a code comment — and wait for the application to pull it into context. That is what makes it dangerous, since the victim is using the tool normally and honestly. Direct injection, by contrast, requires the person at the keyboard to be the attacker.

Explainers using this term

• LLM Security — Prompt Injection and How to Actually Defend Against ItNot What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection