Secure model deployment and adversarial testing

The brief

Attackers will try to evade or abuse any security model. Build a detection model (phishing URLs or intrusion detection), deploy it as an API with authentication, rate limits and logging, then attack it: craft inputs that evade it and abuse the API, and harden it. Good looks like: a deployed, authenticated API, an adversarial test report, and measured improvement after hardening.

What you will build

1. Build and evaluate the detection model
2. Deploy it with authentication, rate limits and logging
3. Craft evasion attempts and measure the drop
4. Harden the model and the API
5. Monitor abuse and drift
6. Write a security test report

Free, open data to build it with

• PhiUSIIL phishing URL dataset ↗ UCI Machine Learning Repository
• UNSW-NB15 ↗ UNSW Canberra

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.