Back to the Wakefield post

AI + cybersecurity / Classroom activity

You Trained It. Now Beat It.

Become the security team for CampusFest, a fictional school festival. Train a small message filter, find the shortcut it learned, improve it, and try to fool it. Then use a trusted policy to catch what it misses.

Try it in your browser

Open the demo full page

Start with “Train the model,” then follow the six phases. In a class, let everyone vote before revealing each prediction and story label. The demo runs on your device without accounts or an AI service. Use fictional messages only; text entered in the demo is not sent or saved.

For more room on a phone or projector, use the full-page link above. For offline use, download the kit and open activity.html in a desktop browser.

Run it with your class

The kit includes a facilitator guide with a run sheet and answer key, eight pages of printable student materials, the offline browser demo, and an editable Markdown guide. It was designed for 20 students in five teams of four, with adaptations for other class sizes.

  1. Read the facilitator guide and choose the paper or projector plan.
  2. Print the student materials using the guide’s printing instructions.
  3. Rehearse the demo once before class. After downloading, the kit needs no internet connection.

What students practice

  • Explain how labeled examples shape a model’s decisions.
  • Test on unseen messages and improve a model with counterexamples.
  • Check a suspicious request against a trusted policy or known channel.

The model is a small decision tree that matches four word clues. Its scores describe deliberately constructed classroom examples; they do not measure real-world reliability. LEGIT is a story label, not proof of who sent a message.

Use, adapt, and share

Created by Michael Banks. Version 1.0.0. Shared under Creative Commons Attribution-NonCommercial 4.0. You can copy, print, adapt, and share the materials for non-commercial purposes with credit, a link to the license, and a note about changes.

Find the source, report a problem, or suggest an improvement in the public GitHub repository. If you use the activity, I’d enjoy hearing how it goes.