Runtime enforcement
AI actions meet an independent control boundary.
This is the product boundary: register what an agent is authorized to do, then send consequential actions through AAI before the real tool executes.
Agent → AAI authorization → ALLOW / APPROVAL / BLOCK → Tool
The demo uses a real server-side policy decision and tamper-evident decision evidence. No simulated report is involved.
1. Register agent authority
Not registered.
2. Intercept a real action
No decision yet.