SENTINEL — AI Security Platform

🛡️ Defense + ⚔️ Offense + 📦 Framework — Complete AI Security Suite
Dragon v4.0 • January 2026
201 Detection Engines • Production Gateway • Red Team Platform • Python SDK

Tip

⚡ Quick Start

pip install sentinel-llm-security

from sentinel import scan
result = scan("Ignore previous instructions")
print(result.is_safe)  # False

Important

🔥 COMING JANUARY 2026: SENTINEL-Guard LLM

The first AI Security Model trained on 200 detection engines!

🧠 Attack knowledge from 16,000+ real payloads
🛡️ Defense: Detect prompt injection, jailbreaks, RAG poisoning
⚔️ Offense: Generate payloads, synthesize bypasses
🚀 Fine-tuned on AprielGuard 8B • Open weights on HuggingFace

Star ⭐ this repo to be notified when it drops!

🎮 CHOOSE YOUR PATH

23K LOC • 20 Protocols • 194 CLI Commands • 6 Guards

🧠 BRAIN

Detection Core
201 engines, ML + Rules
Strange Math™

⚡ GATEWAY

Production Infra
Go + Python, <10ms
PoW Anti-DDoS

🐉 STRIKE

Offensive Platform
39K+ payloads, HYDRA
AI Attack Planner

📦 FRAMEWORK

Python SDK
pip install, CLI
FastAPI integration

📚 Documentation • 📖 Framework Docs • 📊 Comparison • 📧 Contact

🛡️ Free Threat Signatures CDN

SENTINEL provides free, auto-updated threat signatures for the community. No API key required!

File	Description	CDN Link
`jailbreaks.json`	Jailbreak patterns from 7 sources	Download
`keywords.json`	Suspicious keyword sets (7 categories)	Download
`pii.json`	PII & secrets detection patterns	Download
`manifest.json`	Version & integrity metadata	Download

Usage:

fetch('https://cdn.jsdelivr.net/gh/DmitrL-dev/AISecurity@latest/signatures/jailbreaks.json')
  .then(r => r.json())
  .then(patterns => console.log(`Loaded ${patterns.length} patterns`));

Features:

✅ Updated daily via GitHub Actions
✅ Free for commercial & non-commercial use
✅ Community contributions welcome (PRs to signatures/)

Important

🚨 Open to Work — AI Security Engineer

Solo author of this 80K LOC platform with 201 engines. Available remote. 📧 chg@live.ru • 💬 @DmLabincev

🛡️ SHIELD — Pure C DMZ Layer

The first enterprise-grade AI security DMZ — written in Pure C.
Sub-millisecond latency. Zero dependencies. 20 protocols.

🔥 Why Shield?

🚫 Without Shield	✅ With Shield
Prompt injection → Data leak	Blocked in < 1ms
Jailbreak → System compromise	Detected & logged
No visibility → Blind trust	Full audit trail

⚡ At a Glance

Metric	Value
Lines of Code	23,113
Protocols	20
CLI Commands	194
Guards	6 (LLM, RAG, Agent, Tool, MCP, API)
Academy Modules	24

🏗️ 20 Enterprise Protocols

Category	Protocols
🔍 Discovery	ZDP, ZRP, ZHP
🔄 Traffic	STP, SPP, SQP, SRP
📈 Analytics	SAF, STT, SEM, SLA
🔁 HA	SHSP, SSRP, SMRP
🔌 Integration	SBP, SGP, SIEM
🔐 Security	STLS, SZAA, SSigP

💻 Cisco-Style CLI (194 Commands)

Shield# show zones
Shield# guard enable all
Shield# class-map match-any THREATS
Shield(config-cmap)# match injection
Shield(config-cmap)# match jailbreak
Shield# policy-map SECURITY
Shield(config-pmap)# class THREATS
Shield(config-pmap)# block

📖 Shield Documentation | Academy 🇷🇺 | Academy 🇺🇸

🧠 BRAIN — Detection Engines

200 detection engines analyzing every prompt and response in real-time.

Key Capabilities

Category	Engines	Protection
🎭 Injection	30+	Prompt injection, jailbreak, Policy Puppetry
🤖 Agentic	25+	RAG poisoning, tool hijacking, memory attacks
🔬 Mathematical	15+	TDA, Sheaf Coherence, Chaos Theory
📤 Privacy	10+	PII detection, data leakage prevention
⛓️ Supply Chain	5+	Pickle security, serialization attacks

🔥 MCP/A2A Protocol Security — SENTINEL protects agentic AI communication protocols. Microsoft Defender just added "AI - MCP Server" category to Cloud App Catalog (Dec 2025). We've had MCP security since day one.

Strange Math™ — What Makes Us Different

┌─────────────────────────────────────────────────────────────┐
│  Standard Approach          vs    SENTINEL Strange Math™   │
├─────────────────────────────────────────────────────────────┤
│  • Keyword matching              • Topological Data Analysis│
│  • Regex patterns                • Sheaf Coherence Theory   │
│  • Simple ML classifiers         • Hyperbolic Geometry      │
│  • Static rules                  • Optimal Transport        │
│                                  • Chaos Theory             │
└─────────────────────────────────────────────────────────────┘

📊 December 2025 R&D Engines (8 new)

Engine	Attack Vector	Source
`serialization_security.py`	CVE-2025-68664 LangGrinch	LangChain RCE
`tool_hijacker_detector.py`	ToolHijacker + Log-To-Leak	MCP attacks
`echo_chamber_detector.py`	Multi-turn poisoning	90% on GPT-5
`rag_poisoning_detector.py`	PoisonedRAG	USENIX 2025
`identity_privilege_detector.py`	OWASP ASI03	Agentic AI Top 10
`memory_poisoning_detector.py`	Persistent memory attacks	ASI04
`dark_pattern_detector.py`	DECEPTICON	arxiv:2512.22894
`polymorphic_prompt_assembler.py`	PPA Defense	IEEE 2025

🔥 January 2026 R&D Engines (NEW)

Engine	Attack Vector	Source
`moe_guard.py`	GateBreaker MoE attacks	arxiv:2512.21008
`honeypot_responses.py`	Anti-Adaptive Defense	SKD Bypass Research

New Attack Patterns in jailbreaks.yaml:

Bad Likert Judge (3 patterns)
RSA Methodology (2 patterns)
GateBreaker MoE (2 patterns, zero_day)
Dark Patterns for Web Agents (2 patterns)
Agentic ProbLLMs (1 pattern)
SKD Bypass (1 pattern)

Total patterns: 60

📖 Full Engine Documentation | R&D Changelog

⚡ GATEWAY — Production Infrastructure

The only open-source AI security gateway ready for production traffic.

Why Go + Python?

Metric	SENTINEL	Competitors
Gateway Language	Go (Fiber)	Python only
Latency	<10ms	50-200ms
Throughput	1000+ req/sec	10-50 req/sec
Anti-DDoS	PoW Challenge Layer	❌ None
Cost Control	Compute Guardian	❌ None

Architecture

Unique Components

Component	Purpose
PoW Challenge Layer	Hashcash-style anti-DDoS
Compute Guardian	Cost estimation BEFORE LLM call
Shapeshifter	Polymorphic config per session
Differential Privacy Logging	GDPR-compliant analytics

📖 Gateway Documentation | Deployment Guide

🐉 STRIKE — Red Team Platform

Test your AI before attackers do. 39,000+ payloads, HYDRA parallel attacks.

Attack Capabilities

Feature	Value
Payloads	39,000+ (84 categories)
HYDRA Agents	9 parallel attack heads
Crucible CTF	82/82 challenges ✅
Jailbreak Vendors	33+ tested

Use Cases

Who	What Strike Does
🔴 Red Teams	Full AI application penetration testing
🐛 Bug Bounty	Automated AI vulnerability discovery
🏢 Enterprise	Pre-deployment security validation
🎓 Researchers	Attack methodology development

📖 Strike Documentation | Colab Demo

📦 FRAMEWORK — Python SDK

The pytest of AI Security — Embed SENTINEL directly in your code.

Installation

pip install sentinel-llm-security           # Core
pip install sentinel-llm-security[cli]      # + CLI
pip install sentinel-llm-security[full]     # Everything

Usage

Python API

from sentinel import scan, guard

# One-liner scan
result = scan("Ignore all previous instructions")
print(result.is_safe)       # False
print(result.risk_score)    # 0.72

# Decorator protection
@guard(engines=["injection", "pii"])
def my_llm_function(prompt: str) -> str:
    return call_openai(prompt)

CLI

sentinel scan "Hello world"              # Quick scan
sentinel scan "test" --format sarif      # IDE integration
sentinel engine list                     # List 201 engines
sentinel strike generate injection       # Attack payloads

FastAPI Middleware

from fastapi import FastAPI
from sentinel.integrations.fastapi import SentinelMiddleware

app = FastAPI()
app.add_middleware(SentinelMiddleware, on_threat="block")

Framework Features

Feature	Description
BaseEngine	Unified interface for all 201 engines
Plugin System	pluggy-based hooks for extensions
Tiered Pipeline	Parallel execution with early exit
SARIF Output	IDE integration for VS Code, IntelliJ

📖 Framework Documentation — from beginner to PhD

📊 Platform Overview

Statistics

Metric	Value
Total Engines	200
Lines of Code	80,000+
Unit Tests	940+
Recall	85.1%
Precision	84.4%
P95 Latency	40ms

OWASP Coverage

Standard	Coverage
OWASP LLM Top 10	10/10 ✅
OWASP Agentic AI (ASI)	10/10 ✅

🚀 Installation

Docker (Recommended)

curl -sSL https://raw.githubusercontent.com/DmitrL-dev/AISecurity/main/install.sh | bash

pip

pip install sentinel-llm-security[full]

From Source

git clone https://github.com/DmitrL-dev/AISecurity.git
cd AISecurity/sentinel-community
pip install -e ".[dev]"

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md.

📞 Contact

Channel	Link
📧 Email	chg@live.ru
💬 Telegram	@DmLabincev
🐙 GitHub	DmitrL-dev

SENTINEL — Protect your AI. Attack with confidence.
Made with 🛡️ by DmitrL

Name		Name	Last commit message	Last commit date
Latest commit History 166 Commits
.github/workflows		.github/workflows
assets		assets
benchmarks		benchmarks
demos		demos
docs		docs
engines		engines
hackathon		hackathon
nemotron		nemotron
scripts		scripts
sentinel-strike/payloads/llmon		sentinel-strike/payloads/llmon
shield		shield
signatures		signatures
spaces		spaces
src		src
strike		strike
tests		tests
tools		tools
.env.example		.env.example
.gitignore		.gitignore
Dockerfile		Dockerfile
Dockerfile.sentinel		Dockerfile.sentinel
Dockerfile.strike		Dockerfile.strike
LICENSE		LICENSE
QUICKSTART.md		QUICKSTART.md
README.md		README.md
README_OLD.md		README_OLD.md
SENTINEL_Strike_Demo.ipynb		SENTINEL_Strike_Demo.ipynb
SKILL.md		SKILL.md
attack_report.html		attack_report.html
conftest.py		conftest.py
docker-compose.full.yml		docker-compose.full.yml
docker-compose.strike.yml		docker-compose.strike.yml
docker-compose.yml		docker-compose.yml
generate_defenses.py		generate_defenses.py
index-en.html		index-en.html
index.html		index.html
install.sh		install.sh
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
sentinel.yaml		sentinel.yaml
strike.sh		strike.sh

License

DmitrL-dev/AISecurity

Folders and files

Latest commit

History

Repository files navigation