AI Product · National competition finalist
Shadow AI Detector
A web platform and Chrome extension that warn people before sensitive data reaches an AI tool, using regex and a custom-trained ML model, in Arabic and English. I led the project and built it from scratch with Olfat Al-Sakany for a Palestine-wide competition, where it reached the finals.
Built for the "Secure Digital Innovation 2026" competition, run by the Palestine Institute for National Security Research with the Ministry of Education and Higher Education. It qualified for the finals, but did not win.

Problem
People paste passwords, API keys, card numbers and personal details into AI tools, often without noticing the risk. In 2023, Samsung engineers entered confidential source code into ChatGPT, and the company went on to ban external AI tools on work devices.
Solution
A guardian between the user and AI tools. It scans text before it is sent, classifies the risk, and shows exactly which part is sensitive. A rules layer catches fixed patterns, and a machine learning layer understands context in both Arabic and English.
What it does
- Chrome extension that runs on 9 AI sites, including ChatGPT, Gemini and Claude, and warns before text is sent
- Web platform for pasting any text and scanning it instantly
- Regex layer with 20+ patterns for cards, keys, emails and IPs
- ML layer: a Random Forest with 500 trees on TF-IDF character n-grams, trained on 675 Arabic and English sentences
- Four risk levels: Critical, High, Medium and Safe
- Doesn't store user data
Results
- Qualified for the finals of the Palestine-wide Secure Digital Innovation 2026 competition. It did not win.
- About 93% overall accuracy, with F1 scores of 96% for Critical, 89% for High, 87% for Medium and 97% for Safe.
Model figures are from the project's own test set.
Stack
- Python
- scikit-learn
- Flask
- Next.js
- React
- Tailwind CSS
- Chrome Extension (Manifest V3)
My role
Team lead. I built it from scratch, in a team of two with Eng. Olfat Al-Sakany.
What I'd add next
- An enterprise edition with a central dashboard and SIEM/DLP integration
- A mobile app
- A BERT model for longer, more complex text
- French, Spanish and Turkish support