How do AI agent permissions work?

AI agent permissions should be action-level. Arc maps each app action to allow, ask, or block so a user can control what the agent may do.

What is allow / ask / block?

Allow runs the action automatically, ask pauses for user approval, and block rejects the action before it reaches the app.

Permissions

Arc uses allow, ask, and block so users can understand exactly what an agent can do inside an app.

Decision	Meaning	Example
allow	The action runs automatically.	read_email, read_calendar, read_orders
ask	The user must approve before the action runs.	send_email, create_event, request_return
block	The action is rejected.	delete_email, delete_event, delete_account

App	Allow	Ask	Block
Email	read_email, draft_reply	send_email	delete_email
Calendar	read_calendar	create_event	delete_event
Account and commerce	read_orders	request_return	delete_account

Use allow for read-only or draft-only actions that do not affect other people.
Use ask for actions that send, create, purchase, refund, schedule, or change state.
Use block for destructive or high-risk actions until you have a strong reason to support them.

Permissions should not be permanent. Arc lets users revoke a grant so future action requests are blocked until the user reconnects the agent.

See demo Start building