You are the
AI Officer.
An AI agent asks for access to payment and banking systems. You decide on every single request: allow or deny. Five minutes, no login.
What it is about
Reading the request before clicking yes.
That is the habit approvals fail on in real life. Wave through three harmless requests and the fourth goes through as well.
Three role scenarios
From the routine case to the request that looks identical at a glance and is not.
Pick the platform
Claude, ChatGPT or Gemini. The question stays the same whichever vendor you use.
German and English
Switchable inside the game, so it works in an international team too.
What teams use it for
- As the opener in security awareness training
- As a way into the question of which approvals can be automated at all
- As a self-test before deciding to allow AI agents in production
Runs entirely in the browser. No sign-up, no data collection, no outbound calls.
Start the gameThen let us talk about your real approvals
The game shows the problem. Devctrl solves it day to day: rules that bind before the run, and a log that proves the approval.