RAG Security

Browse prompts in this category.

Prompts

RAG Embedding and Vector Database Security
Harden vector stores (Pinecone, Weaviate, Qdrant, Milvus, Chroma, pgvector) — namespace tenant isolation, RBAC with mandatory filters, distribution drift monitoring, vector-magnet detection, encryption at rest, pinned embedding model versions. Use when configuring or auditing vector databases.
RAG Ingestion and Document Loader Security
Harden RAG ingestion — source allowlists with SHA-256 provenance, format parser hardening (PDF JS off, XXE off, DOCX macros off, CSV formula stripping), injection and obfuscation detection, chunk validation, and update-diff approvals. Use when building document loaders or ingestion pipelines.
RAG Pipeline Security Architect
Design defense-in-depth RAG pipelines against corpus poisoning — pre-ingest injection scanning, tenant-isolated vector stores, mandatory ACL retrieval filters, structural prompt separation, output PII/injection validation, quarantine. Use when architecting end-to-end RAG systems.
RAG Prompt Construction and Output Security
Harden RAG prompt assembly and output — role/XML-delimited instruction separation, defensive data framing, per-section token budgets, source-attribution verification, independent PII/content guardrails, per-user cache isolation. Use when designing RAG prompt templates or response validation.
RAG Retrieval Validation and Access Control
Harden RAG retrieval — database-layer mandatory ACL filters, relevance-score thresholds, diversity and concentration checks, query injection validation, post-rerank re-validation, and consistent filtering across hybrid/multi-index search. Use when implementing RAG retrieval or re-ranking.