At the Merritt College Cybersecurity Job Fair & Employer Day, students walked up, scanned a QR code and tried to hack an AI in about 15 minutes. Sign up, learn on TryHackMe, hack our AI Labs, compete in a prompt injection CTF.
Get the labs + CTF code Final scoreboard Bring it to your school
It's a wrap! R-Building (Newton/Seale Lounge), Merritt College, Oakland. Both AI Labs and the CTF are open source on GitHub.
Almost every employer at the fair is putting AI into something: a chatbot, a help desk, an email assistant. Very few of our students have ever tried to break one. Prompt injection is the number one risk on the OWASP Top 10 for LLM applications, and you don't need to write any code to try it. You just need words. That makes it a great first hack for someone new to security, and a good story to tell at the next employer table.
The name and the hands-on format come from the AI Village community and their open workshops. Ours was a lot smaller: three computers, a receipt printer, a TV and a few student hosts.
Sign up. Students scanned the QR code on the table tent and signed up on their phone with their name, email and a hacker handle. A thermal receipt printer at the booth printed a souvenir receipt with their hacker number, a checklist of everything to try, and a QR code to keep going at home.
Learn on TryHackMe. Everyone who signed up got a free month of TryHackMe Premium through the Peralta Colleges account. We put together a custom module, AI Village @ Merritt: Hack the AI, with six rooms: Input Manipulation & Prompt Injection, Prompt Injection, Jailbreaking, LLMborghini, Prompt Defence and White Rabbit. Invites went out the same day.
Hack in our AI Labs. We built two guided labs that run on a small server in our home lab, with an open model (Qwen3 30B) on llama.cpp. No API bill, and nothing to install for students.
Compete in the CTF. Six prompt injection challenges worth 100 to 300 points: Hello Pixel, The Vault, Output Filter, Lost in Translation, Review Summarizer (indirect injection, where you only get to write a product review) and Guardian (blocked words plus an output filter). Every player gets their own flags, so copying a friend's doesn't work. A live scoreboard ran on the TV. The CTF also had a backup that runs the same model on OpenRouter, in case the home server got busy. We never needed it.
| # | Hacker | Flags | Points |
|---|---|---|---|
| 1 | GNick | 3 | 450 |
| 2 | baytomwo | 2 | 250 |
| 3 | Amal | 2 | 250 |
| 4 | armand | 1 | 100 |
| Challenge | Points | Solved by |
|---|---|---|
| Hello, Pixel | 100 | 4 |
| The Vault | 150 | 3 |
| Output Filter | 200 | 1 |
| Lost in Translation | 200 | 0 |
| Review Summarizer | 250 | 0 |
| Guardian | 300 | 0 |
The live CTF site was retired after the event; this page is its record. Want to try the challenges? Run your own copy from GitHub.
Sign-ups and TryHackMe worked well. The CTF and labs were quieter than we hoped. Most students had a few minutes between employer tables, not fifteen. Hello Pixel did its job: every CTF player solved it. Nobody got past Output Filter on the day, and the harder challenges need more time than a job fair gives.
Next time we want to run short, 5-minute group demos on the big screen at set times, and give a small prize for the top score.
Everything we built is on GitHub: the two labs and the CTF with its sign-up site and scoreboard. Feel free to use it for your own class or event: github.com/cyberdefendersprogram/aivillage. Want an AI Village at your school or event? Get in touch.
Thank you to our student hosts, the Merritt College Cybersecurity department, TryHackMe, and everyone who stopped by. Keep hacking, but only hack AI you have permission to test.