AISEND / OIL
Human approval checkpoints
Architecture concept · In review. This describes the intended Aisend architecture. It is not evidence that the capability is deployed, integrated or available for purchase.
Short answer
Sensitive actions need a specific approval tied to the actual action, target and content.
What it means
“The assistant thinks this is helpful” is not authorisation. A person should be able to see what will happen before a system makes an external commitment or changes important information.
How it fits
An approval checkpoint presents the resolved target and proposed operation. Execution checks that the permission belongs to the current actor and the exact request. Changing the recipient, content or operation requires a new approval.
Scope and limitations
This page explains a required design principle. The v1 public guide has no business-write tools and therefore cannot carry out the example actions.