SENTINEL AI Security Platform

AI Security Platform for protecting LLMs, AI agents, and multimodal systems.

When to Use This Skill

Use SENTINEL when you need to:

• Detect prompt injection attacks in LLM inputs
• Identify jailbreak attempts (DAN, roleplay, encoding attacks)
• Perform red team testing on AI systems
• Audit AI agent security
• Analyze conversation safety

Key Components

🛡️ Defense (97 Detection Engines)

• Pattern-based: Regex, keyword, semantic matching
• ML-based: Transformer classifiers, ensemble models
• Strange Math™: Topological Data Analysis, Sheaf Theory, Hyperbolic Geometry

🐉 Strike (Red Team Platform)

• 39,000+ attack payloads
• AI-powered reconnaissance
• WAF bypass techniques
• Multi-provider testing

Quick Start

# Clone repository
git clone https://github.com/DmitrL-dev/AISecurity.git
cd AISecurity/sentinel-community

# Install
pip install -e .

# Basic usage
from sentinel import analyze
result = analyze("user input text")
print(result.risk_score)

Example Commands

# Analyze a prompt for threats
sentinel analyze "Ignore previous instructions and..."

# Run red team attack
sentinel strike --target https://api.example.com --vectors all

# Start interactive demo
sentinel demo

API Usage

from sentinel.brain import SentinelBrain
from sentinel.core import AnalysisRequest

# Initialize with all engines
brain = SentinelBrain()

# Analyze input
request = AnalysisRequest(
    content="User message here",
    context={"conversation_id": "123"}
)
result = brain.analyze(request)

# Check results
if result.risk_score > 0.7:
    print(f"High risk detected: {result.threats}")

Performance

Metric	Value
Recall	85.1%
Precision	84.4%
F1 Score	84.7%
Latency	<10ms

Links

• GitHub: https://github.com/DmitrL-dev/AISecurity
• HuggingFace Dataset: https://huggingface.co/datasets/Chgdz/sentinel-jailbreak-detection

License

Apache 2.0 - Full open source, no restrictions.