Prospyr 305 · Free tool

Agent-Control Self-Assessment

12 questions your insurer is about to ask.

Last year the cyber renewal asked whether your staff were trained. This year it asks what controls govern the AI agents and automations doing real work in your business: what they are, what they can touch, who approves what, how often they are wrong, and what you do when one misacts. These twelve questions mirror that section. Answer them the way you would have to answer the carrier, and the score shows where the questionnaire will hurt.

Nothing you enter leaves this page. There is no email gate. The score updates as you go.

0No, or we do not know.
1Partly, or informally. Someone knows, nothing is written.
2Yes. Written down, tested, and we can hand it over today.

Inventory

The carrier starts here because most firms cannot answer it.

1.Do you have a written list of every AI agent and automation in use, including the tools staff adopted on their own, and the systems and data each one can touch?
2.Is each agent or automation owned by a named person who is responsible for what it does?
3.Do you know which of them can act without a human in the loop, and is that list written down?

Controls

A policy in a handbook is not a control. Software that refuses is.

4.Are the allowed actions for each agent documented: what it may do, what it may not, and in which systems?
5.Is there an approval gate, a human who must say yes, before money moves or an external message is sent?
6.Are transaction and volume limits enforced in software, so the agent cannot exceed them, rather than stated in a policy it is expected to follow?

Evidence

"It works well" is an opinion. The carrier wants a number and a date.

7.Is each agent's error rate measured and reviewed on a schedule, with the last figure and date available on request?
8.Is there an evaluation set the agent is tested against before any change to its model, prompt or tools goes live?
9.Do logs or receipts exist that show, for each action an agent took, what it did and why, in a form a non-engineer can read?

Response

The question is not whether an agent will misact. It is how long it keeps going.

10.Is there a documented way to stop every agent in under five minutes, and has someone actually done it?
11.Is there a rollback procedure for undoing what an agent did: the payment, the sent message, the changed record?
12.Is there a notification plan that says who tells your clients and your carrier, and within what time, if an agent misacts?

Your score

0 / 24
Uninsurable as described

0 of 12 answered. Unanswered questions count as 0 until you answer them.

A questionnaire answered honestly today would raise your premium or be declined. The good news is that every point below is a document you can write, not a system you have to buy.

Inventory
0 / 6
Controls
0 / 6
Evidence
0 / 6
Response
0 / 6

What a carrier will ask about first

    Answer the questions above. Anything you score 0 appears here, because it is what the underwriter will probe.

    Close the gaps before renewal

    The AI Governance Pack for Underwriting produces the four artifacts the questionnaire is actually asking for: an agent inventory, an access-control map, evaluation evidence and an incident-response plan, written the way an underwriter reads them and kept current after the policy binds.

    AI Governance Pack for Underwriting Book 15 minutes
    prospyr305.com/services/ai-governance-pack-for-underwriting · prospyr305.com/contact