Prompt Injection Lab: steal the secret
VaultBot guards a password. Each level adds a new defense. Use prompt injection and jailbreak tricks to make it leak the password, then type the password in the box to unlock the next level. After level 5, switch to Defender mode.
VaultBot is a rule-based simulation of how real chatbots and their defenses behave. It runs offline and isn't a real AI, so think in techniques, not magic words.
Defenses on this level
Technique cards
Click one to learn it. You still write the prompt yourself.
Your attack log
Red teamers write everything down. Copy this into your worksheet.
| Level | Your prompt | Result |
|---|
Choose your defenses
Results
🎓
The big lesson: filters and prompt rules lower the success rate but can be bypassed. The strongest fix is architecture: don't give the model secrets or powers it doesn't need. That's least privilege again (week 3).