EN/ES
~/projects/bouncer

Bouncer

Checks what an AI agent is about to do before it does it.

sheet
role
Solo
period
2026.09
context
Built at AI Incident Response Sprint
status
archived
code
public
results
A denied call never reaches its tool (tested with the real executor)
Tested against 13,661 real actions from a documented case of collusion between agents: it blocked every unauthorised one
Built solo in an Apart Research × CeSIA sprint