Capability is not authority#
An AI system's ability to perform an action does not establish legitimate authority to perform it. Capability, instruction, access, and authorization are distinct: a system may be able to act, be told to act, or possess credentials without having permission from an accountable principal to make that specific change.
What changes when machines can act#
The distinction becomes consequential when AI can invoke tools, alter repositories or infrastructure, communicate externally, access data, spend resources, publish, or change durable external state. Permission to Act examines accountable principals, delegated purpose, resource and operation scope, action-boundary authorization, escalation, revocation, evidence, and responsibility.
Related essay#
Alignment Is Not Authorization: The Permission Layer Agentic AI Now Needs is the concise essay-length treatment of the distinction this book explores more fully.
Publication and access#
Written by Lynn Walker and published by WinMedia on September 20, 2026.
Print editions#
Paperback and hardcover editions are also published, as reported by the publisher for the week ending October 3, 2026. Direct paperback and hardcover listings are available in the edition links above. The publisher report does not establish an exact first print-publication day.