Skip to main content
Discuss Book a call

Free assessment · quick to answer

AI Governance Assessment

Somebody will ask who answers for your AI: a regulator, a customer or your board. This AI governance assessment shows risk, legal and compliance leads where nobody does, and what to fix first.

Built by XONIK. Nothing you type leaves your browser.

Line illustration for an AI governance assessment: rings of governance around AI systems, with one ring left open.

In brief

This free AI governance assessment shows who is accountable for the AI your organization uses and where the gaps are. You answer questions on accountability, inventory, oversight, change and incidents, suppliers and records. You get a governance band, the thinnest areas with their steps, and a report to download. Nothing here is an audit.

Step 1 of 6: Accountability

0 of 18 answered

  1. Accountability (current step)
  2. Inventory (not yet reached)
  3. Oversight (not yet reached)
  4. Change and incidents (not yet reached)
  5. Suppliers and third parties (not yet reached)
  6. Records (not yet reached)

Accountability

Accountability is whether a named person answers for each AI system, who may approve one for use and whether senior people see how AI is used. Without it a problem has nobody whose job it is to act, and every control below depends on someone doing so.

1. Is a named person accountable for each AI system your organization uses or builds?
2. Who may approve an AI system for use with customer, patient or employee data?
3. Does an executive or a board committee receive a report on the organization's use of AI and its risks?

Inventory

Inventory is the list of AI systems and AI features in use, with the data each touches and the decisions about people it influences. A control can't cover a system nobody knows exists, so everything else here reaches only as far as this list does.

4. Do you know which AI systems and AI features are in use, including those built into products you buy and tools staff chose on their own?
5. For each system on the list, is it recorded what data it uses and which decisions about people it influences?
6. Has each use been screened against the uses that the laws applying to you prohibit outright or regulate more strictly?

Oversight

Oversight is whether a person can review, question or reverse what an AI system produces, and whether quality is checked while it runs. Rules on paper count for little if nobody is placed to catch a wrong output before it reaches someone.

7. When an AI system affects a person, can a human review or reverse the outcome?
8. How is the quality of live AI output checked?
9. Do the people who oversee or rely on a system know what it can get wrong, and may they override it?

Change and incidents

Change and incidents covers what happens when a model, a prompt or a supplier changes under a live system, and what happens when something goes wrong. A system that was sound on the day it was approved can stop being sound without anyone deciding it should.

10. When a model, a prompt or a vendor behind a live system changes, what happens?
11. If an AI system caused harm or a serious error, is there an agreed route for reporting and responding?
12. If the owner judged a system unsafe, could they suspend it the same day without asking anyone?

Suppliers and third parties

Suppliers and third parties is how you choose, contract with and keep watch on those who provide AI to you or use it to deliver your work. Much of the AI an organization depends on is built by others, so your cover reaches only as far as what you've asked of them.

13. Before you buy an AI system or an AI feature, what do you ask the supplier?
14. Do your contracts with AI suppliers say what happens to your data and what the supplier must tell you?
15. Do you know which of your own suppliers use AI to deliver work to you?

Records

Records are the logs and the written decisions that let you reconstruct what a system did and why it was approved. When a regulator, a customer or an auditor asks, an arrangement you can't evidence is, in practice, one you can't rely on.

16. For a system that affects people, could you reconstruct what it was asked, what it produced and who acted on it?
17. Is the decision to approve an AI system recorded, with the reasons and the conditions?
18. If a regulator, a customer or an auditor asked tomorrow how you govern AI, could you hand over the documents within a week?

Answer the 3 questions left on this step to go on.

On this page

What does an AI governance assessment tell you, and where does it stop?

It tells you where decisions about AI have no clear owner and where your cover is thinnest. It won't tell you which laws apply, whether you meet them or how well a system performs, and it isn't legal advice. A band describes arrangements you reported. Whether they work is another question, and a regulator will ask it.

Whose knowledge does each area draw on?

Nobody holds every area in their head, so some questions will send you to a colleague.

AreaWho usually knowsWhat may need looking up
AccountabilityThe risk leadWho is named for each system and who may approve a new one
InventoryThe technology lead and procurementTools you bought, tools you built, and AI switched on inside other products
OversightThe owners of live systemsHow outputs are reviewed and who may override them
Change and incidentsThe engineering lead and the head of securityHow a model or supplier change is handled
Suppliers and third partiesProcurement and the general counsel's officeWhat each contract says about your data
RecordsThe compliance leadLogs, the retention rule and approvals

Should you answer alone?

Answer from what you know and the sitting is short. It gets longer if you stop to find out who approves a new system, and those pauses are the assessment doing its work. Answer once alone, then ask risk, engineering and procurement to do the same. The differences between your answers will tell you more than the score.

Is this the right place to start?

Start here when

  • You need to see where decisions about AI have no named owner before anyone drafts a policy
  • A regulator, a customer or your board may ask how you govern AI

Look elsewhere when

  • You need to know which law applies or whether you comply, which is a question for legal advice
  • A third party has to be able to rely on the result, as with an audit

It's quick because it takes your account on trust. It shows you where to look, not what you'll find there.

What does a governance band such as In place tell you?

The labels describe arrangements, not compliance. In place means the arrangements exist, are written down and are used. In place and rehearsed adds that you've run them on a real or practice case and found they hold. Cover is thinnest in the areas that earned the smallest share of their own points. A high band with one area near the bottom still has a hole, and an inquiry will push there first.

What should you do once you have the result?

  1. 01

    Hand it up

    Give the report to the executive who would have to answer a regulator.

  2. 02

    Name an owner

    Put a name and a date against the first step in each weak area.

  3. 03

    Put it on paper

    Turn each step into a list, an authority, a rule or a route.

  4. 04

    Run it on a made-up case

    Try one arrangement on an invented incident and write down what failed.

Want a second pair of eyes on the gaps this showed?

Tell us which area scored lowest and who answers for it today. A senior person reads it and writes back with where we'd start.