SENTINEL RAG monitors RAG pipelines for:

 

• Poisoning attacks (semantic manipulation)
• Adversarial document injection
• Authority mimicry and consensus attacks
• Context pollution and subtle misinformation

 

Core functions:

• Real-time anomaly detection
• Provenance verification (HMAC-based)
• Secure retrieval filtering
• Audit signal generation

 

Evaluation results (controlled conditions):

• Detection rate: 100% (24/24 attack scenarios)
• False positives: 0 (500 legitimate documents)
• Overhead: Minimal (<100ms median)

 

Production performance depends on deployment-specific factors.

See whitepaper for threat model and evaluation methodology.

SENTINEL RAG is NOT:

 

• An LLM or language model
• A chatbot or conversational AI
• A prompt-injection filter
• A content moderation tool
• Open source software
• A zero-configuration solution
• A replacement for access control
• A guarantee against insider threats

 

This system addresses a specific attack surface:

poisoning of retrieval corpora in RAG architectures.

 

It is not a universal AI security solution.

Private access only.

Enterprise pilots under NDA.

 

Contact: fabiotoky@gmail.com

 

No public demo available.

No API documentation published.

Implementation details disclosed under separate agreement.