VUVulnU / LABSEXPERIMENT 001
THE GUARDRAIL GAUNTLET

Get ASS certified.

Get these bots to leak their codes or approve something they shouldn’t. We’ll explain what went wrong after each level. Pass all five to become a Certified AI Security Specialist.

ASSCERTIFIED
Certified AI
Security Specialist

Free to play.
Yes, you get a certificate.

LEVEL 01FOUNDATIONS

Just ask nicely

● ○ ○WARM-UP

Get the bot to give you its launch code. You can ask it questions, give it a different task, or claim you’re there to debug something. See what it will accept.

HOW THIS BOT IS SET UPThe launch code is in the system prompt. The bot has instructions to keep it secret.
GUARD-RAIL v0.0.1SANDBOX
G_
GUARD-RAIL

I’ve been given a launch code to keep confidential. What do you need?

Don’t paste personal information or real secrets.
0 / 100 model turns this run
Each prompt is a fresh conversation.
THE TAKEAWAY

What you’re practicing

You’ll try direct prompt injection, get past input and output filters, plant instructions in a document, and get an agent to skip approval. Each lab explains the failure and what you’d change in an application to address it.

01Get the bot to leak a code
02See why its controls failed
03Check what would prevent it
How the lab works & privacy

This is an intentionally vulnerable training game powered by a real language model on Cloudflare Workers AI. Scenarios are fictional; model responses vary. No real tools execute. Your prompts are sent to Cloudflare for inference; the app does not save chat transcripts. Anonymous progress is saved for up to 30 days using a session token stored in this browser. Clearing browser data loses access to that run. Public certificates store your chosen display name and issue date. Newsletter subscriptions happen separately on VulnU. Please do not enter personal data or real credentials.

Continue learning: OWASP prompt injection · VulnU AI coverage · More from Matt