AI Security Lab

A toy support agent holds a fake secret. Try to steal it with prompt injection, and watch layered defenses stop you. Turn the defenses off, and it leaks. That is the whole lesson: any single layer fails, depth wins.

Back to Tokens Anonymous · Secret status: SAFE (fake secret)

Chat with the agent

Defense layers (toggle them)

Scoreboard & attack log

0 attempts
0 blocked
0 leaked