Agentic AI Action Scope Document

Submitted by

Name *

Email *

Department

Deployment

Deployment name *

Underlying tool *

Purpose

Scope of authority §6.5.C

Specific actions permitted *

Systems and data the agent may access *

Conditions under which the agent operates

Maximum scope of a single action *

Escalation path when the agent hits a condition outside its scope *

Controls §6.5.G–J

Rollback capability in place? *

Required before any write or execute permission is granted (§6.5.H).

How is every action logged? *

Does it contact external parties?

Agents may not contact customers, vendors, or regulators without human review of each message (§6.5.J).

Proposed review date

Annual at minimum, and immediately on any change to permissions, integrations, or task scope (§6.5.F).

An agentic action outside these boundaries is a Category C incident under §8.4, regardless of whether it caused harm.