Trust Fall: An AI Decision Game

How do you play Trust Fall?

Read each AI request, then choose. Approve as proposed keeps any checks already included in the request. An additional human check adds a reviewer beyond those checks. After each call you see what happened, how the meters moved, and what the best call was.

• Green: approve as proposed. The request goes ahead with the checks it already includes. On a keyboard, press the right arrow or A.
• Amber: require an additional human check. Add a reviewer beyond the checks already in the request. On a keyboard, press the up arrow or H.
• Red: reject as proposed. The request does not go ahead as written. On a keyboard, press the left arrow or R.

Every call moves three meters: business value, trust and team time. Approving risky requests burns trust and cleanup time. Rejecting useful ones burns value. Adding checks to low-risk work burns team time, while a well-placed check saves some rework. If any meter hits zero, the desk is shut down. Make the best call on all six requests in a level for three stars.

From level 4, one request in each level follows from an earlier call in the same level. The request says which call it follows and shows any meter change, so nothing depends on a hidden rule. An unfinished level picks up where you left it in this browser.

What do the levels cover?

Play three levels without signup. Access all 10 free with your email, which also brings an email when a new game is added. Progress, including an unfinished level, saves in this browser, so clearing browser data or switching browsers or devices can lose it.

The weekly challenge picks six requests from the main levels every Monday at 00:00 UTC. It is free for everyone, keeps its own score, and lets you copy your result. The requests are a rotation of existing ones, not new scenarios.

Why is it called Trust Fall?

In a trust fall, you decide what to catch. Every request in the game asks the same thing: should the company approve this use of AI as proposed, add another human check, or not go ahead with it?

The game is built around a common mistake. Teams either approve AI because the demo looked good, or send everything to a reviewer and call it safe. Neither works. Review costs real time, and a reviewer who approves hundreds of outputs a day becomes a rubber stamp. The skill is putting the human check where a mistake would really hurt.