← Back to security

AI Context Poisoning & Prompt Injection

6 questions · advanced

Test your understanding of indirect prompt injection vectors, gradient-based adversarial embedding optimization, control token role hijacking, multi-tenant vector index isolation, dynamic delimiter sandboxing, and dual-LLM privilege boundaries.

0 / 6

How do indirect prompt injection vectors leverage white-on-white text, hidden CSS elements, or invisible metadata in ingested PDFs and HTML documents to compromise RAG pipelines?

Press 1 to 4 to pick an answer