🛡 PhishGuard AI — Generative AI Powered Phishing Detection & Awareness Platform

🚀 Live Demo

Try PhishGuard AI here:

https://phish-guard-ai-theta.vercel.app

---

📸 Screenshots

Phishing Analyzer & Simulator

Analyzer Input	Analyzer Result

Simulator Fields	Simulator Generated Message

---

🎯 Project Overview

PhishGuard AI is a full-stack cybersecurity awareness platform that uses Generative AI + Rule-Based NLP to help users detect phishing messages, understand manipulation tactics, and learn how to protect themselves from social engineering attacks.

The Cybersecurity Problem

Phishing attacks are responsible for 90% of data breaches worldwide. They exploit human psychology — not technical vulnerabilities — making traditional security tools insufficient. Users need:

1. Real-time detection of suspicious messages
2. Plain-language explanations of why something is dangerous
3. Hands-on exposure to phishing tactics in a safe environment

---

✨ Features

1. 🔍 Phishing Message Analyzer

• Paste any email, SMS, or message for instant analysis
• Multi-layer NLP detection: urgency words, sensitive data requests, impersonation patterns, suspicious links
• Risk Score (0–100%) with Safe / Suspicious / High Risk classification
• Highlighted suspicious phrases with color-coded threat levels

2. 🤖 AI Explanation Engine

• Powered by Claude (claude-sonnet-4-20250514)
• Explains why a message is dangerous in plain, non-technical language
• Identifies specific manipulation techniques used by the attacker
• Provides actionable advice for the user

3. ⚡ Phishing Attack Simulator

Six educational attack scenarios:

• 🏦 Bank Account Verification
• 💼 Fake Job Offer
• 🔐 OTP Theft Attack
• 📦 Parcel Delivery Scam
• 🏆 Prize / Lottery Win
• 🏛 Government Impersonation

AI generates realistic example messages and annotates each manipulation tactic used.

4. 🛡 Cybersecurity Awareness Tips

• Context-aware security tips shown after each scan
• Covers OTP safety, domain verification, urgency red flags, 2FA setup

---

🛠 Tech Stack

Layer	Technology
Frontend	HTML5, CSS3, Vanilla JavaScript
UI Theme	Cybersecurity dashboard (dark mode, grid/scan effects)
Typography	Orbitron, Rajdhani, Share Tech Mono
Backend	Python 3.9+ with Flask
NLP Engine	Rule-based keyword + pattern matching
AI Engine	Anthropic Claude (claude-sonnet-4-20250514)
CORS	flask-cors

---

📁 Folder Structure

PhishGuard-AI/
├── frontend/
│   ├── index.html      # Main SPA with 3 pages (Analyzer, Simulator, About)
│   ├── style.css       # Cybersecurity dark theme, animations
│   └── script.js       # Detection engine, AI API calls, UI logic
├── backend/
│   ├── app.py              # Flask API server (routes: /analyze, /simulate, /tips)
│   ├── phishing_detector.py # Rule-based NLP detection engine
│   └── phishing_simulator.py # AI simulation + explanation engine
└── README.md

---

🚀 How to Run

Option A: Frontend Only (No Backend Required)

The frontend includes a built-in detection engine and calls the Anthropic API directly.

# Simply open the frontend
open frontend/index.html
# or serve with:
python -m http.server 8080
# Then open http://localhost:8080/frontend/

Note: For direct AI calls from the frontend, you'll need to set your API key in script.js or use a proxy.

Option B: Full Stack (Flask Backend)

# 1. Install Python dependencies
pip install flask flask-cors anthropic

# 2. Set your Anthropic API key
export ANTHROPIC_API_KEY="your_api_key_here"   # Linux/Mac
# set ANTHROPIC_API_KEY=your_api_key_here       # Windows

# 3. Start the Flask backend
cd backend
python app.py
# Server starts at http://localhost:5000

# 4. Open the frontend
open frontend/index.html
# Update BACKEND_URL in script.js to 'http://localhost:5000'
# Set USE_DIRECT_AI to false in script.js

API Endpoints

Method	Endpoint	Description
GET	/health	Check server status
POST	/analyze	Analyze a message for phishing
POST	/simulate	Generate phishing simulation
GET	/tips	Get all security tips

---

🧠 How AI Is Used

Detection Phase (Rule-Based NLP)

No AI needed for the core detection — a hand-crafted ruleset scores messages based on:

• Keyword patterns (3 severity tiers: high/medium/low)
• Suspicious link patterns (regex matching for URL shorteners, IP addresses, typosquatting)
• Urgency language (deadline threats, action imperatives)
• Brand impersonation (known company name detection)
• Sensitive data requests (OTP, PIN, password, CVV)

Explanation Phase (Generative AI)

Claude generates personalized explanations for each analyzed message. The prompt includes:

1. The original message
2. The detected risk score and indicators
3. Instructions to explain in plain language for non-technical users

Simulation Phase (Generative AI)

Claude generates realistic-but-safe phishing examples for educational purposes. The AI:

1. Crafts a scenario-appropriate phishing message
2. Lists specific manipulation tactics used
3. Provides a psychological breakdown of the attack vector

---

🔒 Security & Ethics

• All phishing simulations are clearly labeled as educational examples only
• No real phishing infrastructure is created
• The tool teaches recognition, not attack execution
• Detection engine works offline — no message data is sent externally during analysis (only for AI explanation)

---

📊 Detection Accuracy

The rule-based engine achieves ~85-90% accuracy on common phishing patterns. The AI explanation layer catches nuanced context that rules miss. For production use, consider integrating:

• PhishTank or OpenPhish API for URL reputation
• A fine-tuned BERT model on phishing datasets
• Email header analysis (SPF/DKIM/DMARC verification)

---

🚀 Future Improvements

• Integration with phishing URL reputation APIs (PhishTank, OpenPhish)
• Machine learning model for advanced phishing detection
• Email header analysis (SPF, DKIM, DMARC verification)
• Browser extension for real-time phishing protection

📄 License

MIT License — Free for educational and research use.

---

Built with 🛡 for cybersecurity awareness education