Skip to content

RAG Security Lab V2 — Ingestion Integrity & Poisoning Defense

Latest

Choose a tag to compare

@Nameless8243 Nameless8243 released this 02 Dec 23:37
42510d6

🛡️ RAG Security Lab V2

A multi-layer defensive lab demonstrating enterprise-grade security controls for securing RAG ingestion pipelines.


🔹 Features

  • Lineage & integrity verification (hash + HMAC signature)
  • Semantic anomaly detection (embedding-based poisoning & drift)
  • Multi-stage quarantine workflow
  • Tamper-evident audit log (hash-chain)
  • Full adversarial attack simulation
  • Cleanup utility for deterministic runs

🔹 What’s New in V2.0.0

  • Switched to real semantic embeddings (MiniLM-L6-v2)
  • Added semantic centroid baseline
  • Added weighted anomaly scoring (alpha)
  • Added enterprise-grade output formatter
  • Added reset_data.py cleanup tool
  • Improved documentation & architecture diagrams
  • Restructured folder layout for testability
  • More realistic, multi-stage poisoning simulations

🔹 How to Run

python3 -m simulations.full_attack_simulation


🔹 Cleanup

python3 simulations/reset_data.py


🔹 License

  • MIT License — educational & research purposes only