Free reference · Annexwright
The AI questions in enterprise security questionnaires — and what each answer has to rest on
Enterprise security reviews and vendor procurement questionnaires added an AI section in 2025–26. These are the twenty questions that actually turn up, grouped the way questionnaires group them. The second column is the part that catches people out: each answer only survives follow-up if there is a document behind it.
Written by Yosef Dabbit, Annexwright · Published 2026-09-23 · Last reviewed 2026-09-23 · Edition 2026.1
If a questionnaire is open on your desk right now
Work through the twenty questions below and mark every one where you could not attach the supporting document today. That list is your actual gap. The documents that close it are in Essentials, $149 — including the answer bank these examples come from.
Why the second column matters more than the first
Most vendors can improvise an answer to any single question here. The problem arrives one email later, when the reviewer asks to see the thing the answer described. "Yes, we maintain an AI inventory" becomes "please attach it." An answer without an artifact behind it reads as a claim, and claims slow deals down.
So the useful preparation is not drafting answers. It is knowing which document each answer points at, and having that document exist before the questionnaire arrives. That chain — question, answer, artifact, record — is set out in full here.
The twenty questions
1. Governance
| The question | What your answer has to rest on |
|---|---|
| Do you have an AI governance policy approved by senior management? | a signed AI governance policy |
| Who is accountable for AI governance? | a named accountable owner, not a department |
| Are you ISO/IEC 42001 certified or aligned? | a signed AI governance policy; an implementation plan |
| Do you maintain an inventory of AI systems? | an AI system inventory covering third-party models and AI features inside SaaS tools |
| Which regulations govern your AI use and how do you comply? | a per-system regulatory applicability assessment |
2. Risk and impact management
| The question | What your answer has to rest on |
|---|---|
| How do you assess AI risk? | a scored AI risk register; an impact assessment with prohibited-practice and high-risk screens |
| Do you test for bias or fairness? | an impact assessment with prohibited-practice and high-risk screens |
| Do you use AI for decisions about individuals? | a documented human oversight procedure |
| How do you manage AI incidents? | an AI incident response procedure and log |
3. Data
| The question | What your answer has to rest on |
|---|---|
| Is customer data used to train AI models? | a generative AI usage standard; your vendor contract terms |
| What data is sent to third-party AI providers? | a generative AI usage standard; a data governance procedure for AI |
| Which AI sub-processors do you use? | a published sub-processor list |
| How do you ensure data quality and provenance for AI? | a data governance procedure for AI |
4. Security
| The question | What your answer has to rest on |
|---|---|
| How do you protect against prompt injection and data exfiltration? | an impact assessment with prohibited-practice and high-risk screens; a generative AI usage standard |
| Are AI tools covered by your access control and logging? | a generative AI usage standard |
| Do you monitor AI systems in production? | a documented human oversight procedure |
5. Transparency and people
| The question | What your answer has to rest on |
|---|---|
| Do you disclose AI use to end users? | transparency and disclosure notices |
| Can your customers turn off AI features? | your product documentation |
| Are your staff trained on responsible AI? | AI literacy training records |
| How do you vet AI vendors? | a vendor due diligence questionnaire |
What a supported answer looks like
Three rows from AW-E17, the Enterprise AI Questionnaire Answer Bank, exactly as they ship. The fourth line in each is the part reviewers actually escalate to: the record that proves the artifact is live rather than shelved.
The caution that ships with the document. AW-E17's own instruction is blunt: replace bracketed facts with your reality — never claim a control you do not operate. A starting answer is a pattern to edit, not a statement of fact about your company. An answer that overstates what you do is worse than an honest gap, because the follow-up request is where it fails.
Where to go next
How this page was checked. Regulatory dates on this page were taken from the official texts and from the European Commission, the Illinois General Assembly, Texas Legislature Online, the Colorado General Assembly, the California Privacy Protection Agency, the New York State Senate and NYC DCWP, and were last reviewed on 2026-09-23. Where a date is an applicability or compliance deadline rather than an entry-into-force date, the page says so. Regulation changes; check the primary text before relying on any date for a decision. Nothing here is legal advice.