HD Field Guide 11
RAG security: treat retrieval as untrusted
Protect retrieval-augmented generation from poisoned documents, indirect prompt injection and excessive data access.
Control areaApplications
AudienceDevelopers & security teams
Reading time8 min
Reviewed11 August 2026
RAG brings external content into model context. That content can be wrong, malicious, over-broad or unauthorised.
Protect the retrieval boundary
Enforce document permissions before content reaches the model.
Retain provenance so analysts can verify output.
Separate content from authority
Retrieved content must not override system policy or approvals.
Validate citations and output before downstream use.
Before you close the guide
- Per-user retrieval authorization
- Provenance retained
- Indirect injection tested
- Execution isolated
Continue with primary guidance
Defensive guidance boundary
This guide provides general educational information. Adapt it to your organisation’s policies, contracts, legal obligations and incident process. For an active incident, use trusted professional and official support channels.