The Security Questionnaire Has a New Page

  1. THE DEAL THAT SECURITY STOPPED Users loved it Security said no They loved it. Security said no. The questionnaire has a new page. Taskeen taskeen.leapora.tech 1 / 4 (c) Leapora AI, Inc. All rights reserved. Siva Kowsika.
  2. WHAT THE NEW PAGE ASKS Not whether it can be broken into BY SOC 2 encryption access control pen testing BY NOTHING who is accountable what record exists how long it is kept can we get it out Taskeen taskeen.leapora.tech 2 / 4 (c) Leapora AI, Inc. All rights reserved. Siva Kowsika.
  3. WHY THE ANSWERS ARE HARD These are architectural, not editorial. Retention you did not design cannot be added by policy. Taskeen taskeen.leapora.tech 3 / 4 (c) Leapora AI, Inc. All rights reserved. Siva Kowsika.
  4. WHY YOUR BUYER IS ASKING The organisations buying are asking because somebody is asking them. Everybody in that chain answers upward to someone who wants it in writing. Taskeen taskeen.leapora.tech 4 / 4 (c) Leapora AI, Inc. All rights reserved. Siva Kowsika.

A deal can be won in every conversation with the people who will use the product and lost to a document filled in by somebody who never saw it.

Anyone who has sold to a large organisation has had this week. The users are advocates, the business case is signed off, and the thing that stops it is a security questionnaire returned with gaps.

The questions on the new page

Who is accountable for what your AI features do. What record exists of a decision one of them influenced. How long is it kept. Can we get it out, in a form our regulator will accept, after you and we have stopped working together.

These are not security questions in the old sense. Nothing in a penetration test or a SOC 2 report answers them, because they are not about whether the system can be broken into.

They are about whether the system can account for itself, which is a different property and one that most products were never asked to have.

Why the answers turn out to be architectural

Vendors meeting this page for the first time reach for the usual instruments and find none of them fit.

Retention you did not design cannot be added by policy. Writing "we retain decision records for 24 months" in a trust centre does not make it so if the underlying table is pruned by a job somebody wrote in 2021 for performance reasons.

A record that was a byproduct of how your database works cannot be made into evidence by writing a page about it. Evidence has to be the thing the system was trying to produce, not a side effect of how it happened to store state.

And provenance for an AI-influenced decision cannot be reconstructed for decisions already made. Whatever the model was shown, and whatever a human did with the output, either got recorded at the time or did not exist by the time anybody asked.

Why the buyer is asking

It is easy to read the new page as procurement theatre, and easy to be wrong about that.

The organisation sending it is asking because somebody is asking them. A bank's vendor management team adds a question because the regulator added one to their supervisory letter. A hospital trust adds one because its own board asked how AI-assisted triage decisions are evidenced.

Everybody in that chain is answering upward to someone who will want it in writing. Which means the question does not go away if a vendor stalls, and it does not get softer in the next renewal cycle.

It also means the buyer is not looking for reassurance. They are looking for something they can hand on.

Procurement becomes the place this gets priced

For most of the last decade, security posture was a gate: you passed or you did not, and passing was worth nothing beyond proceeding.

Accountability is behaving differently, because the answers differ by degree rather than by pass and fail. A vendor who can export a complete decision record in a readable format is doing something materially more valuable for the buyer than one who can produce logs on request, and the buyer's own regulator will notice the difference.

That turns a compliance checkbox into a commercial position, and it does so quietly, in the part of the sale that product teams are usually furthest from.

The vendors who notice early get to design for it. The ones who notice when a questionnaire arrives get to explain, in writing, to somebody with no reason to be generous.

The week the questionnaire arrives

It is worth picturing the actual week, because it is where the cost lands and it never appears in a plan.

A product manager is asked to answer eleven questions by Friday. She does not know the answers, so she asks engineering. Engineering does not know either, not because they are careless but because nobody has ever needed the system described in those terms, and the honest answer to most of them is "it depends what you mean".

What follows is two weeks of archaeology: reading retention jobs, working out what the audit table actually contains, discovering that one field everyone assumed was populated has been null since a refactor. Then somebody writes a careful paragraph that is true and does not say yes.

The deal does not die. It slips a quarter, and it slips again at the next customer, and the pattern repeats until somebody treats it as a product gap rather than a paperwork one.

What answers the page

Who accepted which obligation. What the system did. What a person decided. Held for a stated period and exportable in a form the other side can check.

That last clause carries more weight than it looks. Exportable in a form the other side can check means the buyer can verify it without your cooperation, which is the only version of evidence that survives the relationship ending.

None of it can be added in the week the questionnaire arrives. That is the whole difficulty and also the whole opportunity: a property that cannot be retrofitted is a property a competitor cannot acquire in a sales cycle either.

Questions

What are buyers asking about AI in security questionnaires?

Who is accountable for what the AI features do, what record exists of a decision one influenced, how long it is kept, and whether it can be exported in a form a regulator will accept after the contract ends. None of these are answered by a penetration test or a SOC 2 report.

Why can vendors not answer the AI accountability questions?

Because the answers are architectural. Retention that was not designed cannot be created by policy, a record that is a byproduct of database behaviour is not evidence, and provenance for decisions already made cannot be reconstructed after the fact.

Is the AI page on a security questionnaire just procurement theatre?

No. The organisation asking is answering upward to somebody who wants it in writing, usually a regulator or a board. That means the question does not soften at renewal and the buyer needs something they can hand on, not reassurance.

What does exportable evidence actually mean?

That the counterparty can verify the record without the vendor's cooperation. It is the only form of evidence that survives the relationship ending, which is precisely the scenario the question is asked about.

Taskeen records who was told, who acknowledged, what they did, and what proves it. See how it works.