Skip to main content

Sentinel AI Algorithms

1. Overview (What does this do?)​

The Sentinel AI Algorithms module provides the advanced threat detection capabilities for Ferrox-Py. By integrating ai_guardrails.py and markov_sequence.py, it elevates the Python security layer beyond basic RBAC, offering intelligent, context-aware protection against LLM Prompt Injection and anomalous bot-like behavior.

2. Philosophy (Why does it exist?)​

Modern web applications and AI-driven platforms face sophisticated attacks that traditional static rules (like rate limiting) cannot catch. This module exists to provide a Zero-Trust intelligent mesh, ensuring that every payload and request sequence is probabilistically evaluated for malicious intent before it reaches the core CQRS bus.

3. Target Audience (Who is it for?)​

This module is designed for DevSecOps engineers and backend developers building AI integrations, financial services, or any high-stakes API in Python where automated bots and prompt injections are a critical threat.

4. Architecture (How does it work?)​

The algorithms are instantiated as core security dependencies inside the ferrox_py_auth.security.sentinel package:

  • AI Guardrails (ai_guardrails.py): Intercepts textual payloads and evaluates them against known injection patterns and dimensional risk models (calculating a dynamic risk_score).
  • Markov Sequence Detector (markov_sequence.py): Tracks the sequence of actions a user takes and evaluates it against a trained Transition Matrix. If the transition probability drops below a critical threshold, it indicates non-human (bot/scraper) behavior.

5. Installation / Setup​

These algorithms are natively included within the ferrox-py-auth package. Ensure it is installed in your virtual environment:

pip install ferrox-py-auth

6. Quickstart (Usage)​

You can inject these detectors directly into your FastAPI dependencies or middleware:

from ferrox_py_auth.security.sentinel.ai_guardrails import AIGuardrails
from ferrox_py_auth.security.sentinel.markov_sequence import MarkovSequenceDetector

guardrails = AIGuardrails(block_threshold=0.85)

# Checking a payload before processing
if not guardrails.check_payload(user_input):
    raise HTTPException(status_code=403, detail="Payload blocked by [Sentinel](/docs/ferrox-py/security/sentinel-algorithms) AI Guardrails")

7. Ecosystem Integration​

These algorithms are designed to sit perfectly at Layer 3 of the Ferrox Onion Pipeline (Sentinel Threat Engine). When a Markov Sequence or AI Guardrail is triggered, the system automatically logs the event to the Observability module and can dynamically invoke MTD (Moving Target Defense) strategies to ban the offending IP across the entire distributed cluster.