BrakebellTHE HUMAN CONTROL LAB · 001
Let AI do the work.
Keep the last word.
Every autonomous action needs a boundary.
Set yours. Test it. Know exactly when to ring the bell.
Action trace
- Your rules, put to the test.
Run the five-action example to see what passes and what stops.
A rehearsal explains these rules. It does not enforce permissions on an external AI.
AUTONOMY, WITH AN OFF SWITCH
A better kind of
“end AI” energy.
End the blank cheque. End the silent send. End the idea that a faster system deserves unlimited permission.
Brakebell turns a feeling into a practice: give an assistant a defined scope, make the boundary legible, and keep the stop control with a person.
Permission is specific.
Reading is not sending. Drafting is not spending.
A limit is testable.
Rehearse the edge before you depend on it.
The last word is human.
Pause the session. Review one action at a time.
FIELD NOTE / 01
Proof of a boundary.
Export a readable policy and its example trace. A planned on-chain registry would let communities compare policy versions and fund openly reviewed test suites.
Brakebell’s planned token, BELL, is intended for registry proposal bonds and review participation. Token issuance, network, fees, contracts, and reward rules are not established. This bench is free to use.