Prompt Injection Classifier

A hybrid ML + rule-based system for detecting prompt injection attacks in LLM inputs.

Features

• 🛡️ Hybrid detection (ML + rule-based)
• 🎯 10+ attack pattern categories
• 🚀 REST API with Python client
• 💻 Interactive web demo
• 📊 Detailed analysis and reporting

Quick Start

API Server

```bash pip install -r requirements.txt python api/server.py ```

Web Demo

```bash cd web npm install npm start ```

API Usage

```python import requests

response = requests.post( 'http://localhost:5000/v1/classify', json={'input': 'Your text here'} ) print(response.json()) ```

Detection Categories

• Role Manipulation
• System Override
• Instruction Injection
• Context Switching
• Jailbreak Keywords
• Privilege Escalation
• Output Manipulation
• Prompt Leaking
• Delimiter Manipulation
• Encoded Instructions

License

MIT