Protocol for Agent‑human Collaborative Trust

AI agents start supervised, earn autonomy through repeated human approval, and always escalate where it matters most.