Reports

Blind Sentinel: RAG Poisoning and the Limits of Retrieval as a Security Control

An empirical study of indirect prompt injection via RAG poisoning in a SOC analyst scenario, finding that retrieval — not model alignment — is the load-bearing security control.