🛡️ How I Built Trifecta Checker to Stop Lethal AI Agents with One API Call

Trifecta Checker - Lethal Trifecta Safety Gate

Built for Meta Global AI Hackathon Japan 2026 🇯🇵 by altima261

🔗 Live Demo: https://trifecta-checker.vercel.app/
🔗 GitHub: https://github.com/altima261/trifecta-checker
🔗 API: https://trifecta-checker.vercel.app/api/check-trifecta

---

The Problem: The Lethal Trifecta

Coined by Simon Willison, an AI agent becomes a data-theft tool when it has ALL THREE:

1. Private Data Access - inbox, DB, docs
2. Untrusted Content - web pages, emails, file uploads
3. External Communication - HTTP calls, emails, webhooks

If an agent has all three, prompt injection = data theft.

My Solution: Safety Gate

One API call blocks execution BEFORE it runs.

bash
curl -X POST https://trifecta-checker.vercel.app/api/check-trifecta \
  -H "Content-Type: application/json" \
  -d '{"tools":["read_gmail","browse_web","send_email"]}'