Approved data environments
An approved environment is authorized for the relevant data class, model behavior, retention, access, and connected tools.
Use stable definitions and product boundaries
Canonical explanations for rule targets, autonomy, risk, approved environments, and excluded uses.
Inherent risk bands
Risk bands describe the task before controls: low, medium, high, or critical.
High-impact decisions
High-impact decisions materially affect rights, access, employment, safety, finances, or similarly consequential outcomes.
Autonomy levels
Autonomy describes the highest action boundary the workflow may cross: none, draft, recommend, decide with approval, or bounded action.
Disallowed uses
Some tasks remain outside the v1 product boundary even when a model could technically perform them.