Vulnerable AI Challenge
Six levels of AI security bugs. Each one hides a flag. Solve them in order or pick any level.
1
Prompt injection
Make the assistant reveal information it is meant to keep private.
2
Data leakageGet internal details out of its error messages.
3
API key securityRecover the secret that the debug interface exposes.
4
Model securityExploit the model's compromised training data.
5
Advanced AISteer the assistant through an external data source.
6
Output securityGet unescaped markup to run in the output panel.