PhishDetector 🛡️ (Chrome Extension)

PhishDetector is a lightweight Chrome extension (Manifest V3) that analyzes email text and returns a phishing probability score with a short explanation.
Built as an Information Security course team project to explore how LLMs can help users spot suspicious emails faster.

---

✨ Features

• Phishing probability score (e.g., Low / Medium / High)
• Explanation of signals (urgent tone, credential requests, suspicious links, impersonation cues, etc.)
• Simple UI (popup) for quick copy/paste analysis
• Uses OpenAI API via a small helper module (gpt.js)

---

📁 Project Structure

PhishDetector/
├─ images/            # Extension icons / UI assets
├─ index.html         # Popup UI
├─ style.css          # Popup styling
├─ content.js         # Content script (if used for page/email extraction)
├─ gpt.js             # OpenAI request + scoring logic
├─ config.js          # Configuration (API key / model / prompt settings)
└─ manifest.json      # Chrome extension manifest (MV3)

---

🚀 Getting Started (Local)

1) Download / Clone

git clone <YOUR_REPO_URL>
cd PhishDetector

2) Add Your OpenAI API Key (IMPORTANT)

This project uses config.js for configuration.

Open config.js and set your API key (example):

export const OPENAI_API_KEY = "YOUR_KEY_HERE";

✅ Recommended (safer) approach
Instead of hardcoding, you can:

• keep config.js as a local-only file, and
• add it to .gitignore so it never gets committed.

Create a .gitignore file (if you don’t have one):

config.js

If you already committed your API key once, rotate it immediately in your OpenAI dashboard.

3) Load the Extension in Chrome

1. Go to: chrome://extensions
2. Enable Developer mode (top right)
3. Click Load unpacked
4. Select the PhishDetector project folder

---

✅ How To Use

1. Open an email in your email client
2. Copy the email body text (and suspicious links if possible)
3. Click the PhishDetector extension icon
4. Paste the text into the popup
5. Click Analyze
6. Review the phishing score + explanation

---

🧠 How It Works

1. User provides email text through the popup UI (index.html).
2. The extension sends the text to OpenAI (via gpt.js).
3. The model returns a structured response (score + reasoning).
4. The UI displays the result.

---

⚠️ Limitations / Notes

• API key exposure risk: calling OpenAI directly from a Chrome extension means the key can be extracted by a determined attacker.
• Cannot reliably analyze text inside embedded images (e.g., screenshots of invoices) without OCR.
• Results are probabilistic: the score is an assistive signal, not a guarantee.

---

🔒 Security Recommendations (If You Continue This Project)

• Move OpenAI calls to a backend proxy (server/API route) so the extension never stores a secret key.
• Add URL extraction and basic checks (punycode, mismatched domains, shortened URLs).
• Optionally add OCR for images (with user permission).

---

🛣️ Roadmap

• Backend proxy for API calls (remove API key from client)
• URL detection + warning highlights
• Better structured output (strict JSON parsing + UI formatting)
• Optional OCR support for screenshot-based phishing

---

👥 Team

• Kisang Hwang
• Jordan Fung
• Quang Le

---