AI agent governance checklist: 12 questions

AI agent governance checklist: 12 questions before an AI acts.

Use this before you let an AI system take actions in a regulated workflow. If you cannot answer a question, that is the gap to close first. It follows the approach in AI agent governance: authority, policy, human approval and evidence.

Authority

What exactly is this AI allowed to do, for what purpose and on which systems?
Is there a named owner accountable for each AI system or agent?
Are the limits (amounts, data, systems) written down, not assumed?

Identity

Can you tell which model or agent produced a given recommendation?
Is the version or configuration in use at the time recorded?

Policy

Is the applicable rule or approval threshold applied before an action, not only documented in a register?
What happens when a proposal falls outside policy: is it stopped or escalated?

Human approval

Which actions need a named person to approve?
Can that person approve, reject or override, and is their decision recorded?

Evidence

Could someone who was not there reconstruct the source, the policy, the approval, the action and the result?
Is evidence written as the work happens, not assembled afterwards?
Are overrides and false alarms kept, not discarded?

How to use it

Go through it with the person who owns the workflow and the person who approves in it. Tick what you can evidence today; list what you cannot. A walkthrough can test the gaps against your actual workflow. This is a self-check, not an assessment and not legal advice.