Apppublic

NorthernTribe-Research/ai-phishing-detector

sourceHugging Facemitupdated 7mo agoView on Hugging Face

0likes

App README

![License](https://opensource.org/licenses/MIT) ![Python](https://www.python.org/) ![Docker](Dockerfile) ![Hugging Face](https://huggingface.co/spaces/unit731/ai-phishing-detector)

Enterprise-Grade Neural Threat Detection Engine

A production-ready, multi-modal inference engine designed to detect sophisticated AI-generated phishing, impersonation, and social engineering attacks across Email, SMS, and Social Platforms.

System Overview

This solution leverages Zero-Shot Natural Language Inference (NLI) and Computer Vision (OCR) to identify malicious intent without relying solely on static signature databases. It is designed for rapid deployment in containerized environments (Docker/Kubernetes) or serverless platforms like Hugging Face Spaces.

Core Capabilities

Feature	Description	Technology Stack
Contextual Analysis	Detects coercion, urgency, and fear-mongering tactics.	DeBERTa-v3 (Zero-Shot)
Impersonation Detection	Identifies brand spoofing and executive masquerading.	Stylometry & NLI
Visual Threat Intel	Extracts text from screenshots/images for analysis.	Tesseract / EasyOCR
Threat Intelligence	Validates URLs against risk heuristics and homographs.	Heuristic Engine
Explainable AI	Provides human-readable breakdown of why a message is risky.	Shapley / Attention

Architecture

mermaid

graph TD
    User[User / Client] -->|Text or Image| API[FastAPI / Gradio Interface]
    API --> Router{Input Type?}
    
    Router -->|Image| OCR[OCR Engine]
    OCR -->|Extracted Text| Pre[Preprocessing Pipeline]
    Router -->|Text| Pre
    
    Pre -->|Cleaned Text| NLI[Zero-Shot Classifier]
    Pre -->|Metadata| Heur[Heuristic Engine]
    
    subgraph Inference Engine
        NLI -->|Intent Probabilities| Aggregator
        Heur -->|URL/Urgency Scores| Aggregator
        Aggregator -->|Final Risk Score| Outcome[Risk Assessment]
    end
    
    Outcome -->|JSON / UI| User

Quickstart

Option A: Docker (Recommended)

Run the fully containerized application to ensure environment consistency.

bash

# Build the container
docker build -t unit731/phish-detector .

# Run on port 7860
docker run -p 7860:7860 unit731/phish-detector

Option B: Local Python Environment

bash

# 1. Clone repository
git clone https://github.com/734ai/phish_detector_space_files.git
cd phish_detector_space_files

# 2. Setup Virtual Environment
python -m venv .venv
source .venv/bin/activate

# 3. Install Dimensions
pip install -r requirements.txt

# 4. Launch Application
python app.py

Configuration

The application can be configured via Environment Variables for enterprise integration.

Variable	Description	Default
`HF_TOKEN`	Hugging Face Token for gated models.	`None`
`PORT`	Listening port for the application.	`7860`
`LOG_LEVEL`	Logging verbosity (INFO, DEBUG, ERROR).	`INFO`

Security & Privacy

—Data Handling: This application processes data in-memory. No data is persisted to disk or external databases by default.
—PII Safety: We recommend sanitizing PII (Personally Identifiable Information) before submission, although the model runs locally within the container.
—Container Security: The Docker image is built on python:3.10-slim to minimize attack surface. Run as a non-root user (configured in Dockerfile).

Directory Structure

text

.
├── app.py                 # Application Entrypoint (Gradio/FastAPI)
├── Dockerfile             # Container Definition
├── requirements.txt       # Python Dependencies
├── src/
│   ├── inference.py       # Core Logic & Model Loading
│   ├── ocr.py             # Optical Character Recognition Pipeline
│   ├── preprocessing.py   # Text Normalization & Heuristics
│   └── scoring.py         # Risk Aggregation Logic
└── assets/                # Static Resources (UI Styles, Images)

Contributing

We welcome contributions from the security community. Please follow these steps:

1.Fork the repository.
2.Create a feature branch (git checkout -b feature/AmazingFeature).
3.Commit your changes.
4.Open a Pull Request.

License

Distributed under the MIT License. See LICENSE for more information.

Developed by Unit 731 for a safer internet.