Indirect Prompt Injection
Indirect prompt injection embeds malicious instructions in external content — RAG documents, emails, web pages, or tool outputs — that the model reads without the user typing them directly.
Indirect prompt injection embeds malicious instructions in external content — RAG documents, emails, web pages, or tool outputs — that the model reads without the user typing them directly.