RAG (retrieval-augmented generation)
Also called: retrieval-augmented generation · RAG poisoning
RAG, retrieval-augmented generation, is a pattern where an AI system searches a document store and adds the results to the model's prompt before answering. Everything retrieved becomes model input, so the corpus is an injection and data-leak surface.
How it works
Attackers plant instructions or false facts in documents that will be indexed, or exploit weak access control so a user retrieves documents they should not see. Retrieved text can also carry exfiltration payloads into the answer.
How to defend
Enforce document-level permissions at retrieval time, control who can add content to indexes, mark retrieved text as untrusted, and monitor for planted instructions.
Real cases 0 stories
No story is filed under this term yet. New ones land here automatically.