Step 2: Authorize

How do you control what an AI agent may do and spend?

Give each agent a policy: what it may reach, what it may do alone, what needs a person, and how much it may spend.

  • Access
  • Actions
  • Spend

Request early access

Three tiers

Every action sits in one tier

Acts alone

Small, low-risk tasks.

Acts alone

Needs approval

A named person says yes first.

Needs approval

Never allowed

Blocked, and still recorded.

Never allowed

Limits you set

Four limits

Systems

What it may reach

Actions

What it may do

Spend

What it may use

Roles

Who may change policy

One switch

Stop everything at once

The emergency stop control in MFDIO settings: pauses every active agent and blocks all AI actions at once.
The emergency stop in settings. Only an Owner or Admin can use it.
  • Pauses every active agent
  • Owner or Admin only
  • Recorded on the record

Status

Today Works now

  • Approval by default for risky actions
  • Spend limits: warn, ask or stop
  • Roles decide who sets policy

Roadmap Planned

  • Policy for agents built elsewhere

Quick answers

What happens at the spend limit?

The organization chooses: warn, ask for approval, or stop. Overage is off by default.

Can we start strict?

Yes. Start strict and loosen a limit once the record shows an agent behaving well.

Next: a person decides

Operate sends risky actions to a named approver.

Go to Operate