AI security questionnaires

AI security questionnaire help.

The AI section is where enterprise security reviews stall. We answer it accurately, attach the evidence buyers ask for, and fix the engineering behind any answer that isn’t true yet.

SIG, AI-CAIQ, customEvidence attached
Questionnaire

The AI section, by topic

Training on customer dataYes, no, or opt-in, and how you prove it.
Model providersSubprocessors, regions and retention.
Tenant isolationRetrieval, caches and fine-tunes.
Prompt injectionWhat limits the damage when it works.
Human oversightWho reviews outputs and actions.
Logging and incidentsWhat you record and how you respond.
Six topicsCommon across buyers
Gaps fixed

What the AI section asks, and the evidence that answers it.

The wording changes from buyer to buyer. The topics don’t. For sample answers, read how to answer the AI section of a security questionnaire.

Data use and training

What the buyer is checkingWhether customer data, prompts or outputs train any model, yours or a provider’s, and whether customers can opt out.

Evidence to attachProvider terms and account settings, your own training data policy, and the product setting if there is one.

Model providers and subprocessors

What the buyer is checkingWhich AI vendors process customer data, where, for how long, and under what contract.

Evidence to attachA subprocessor list that includes AI vendors, region settings, and retention terms.

Tenant isolation

What the buyer is checkingThat one customer’s data can’t appear in another customer’s AI answers, search results or fine-tuned model.

Evidence to attachAn architecture diagram, the access checks in retrieval, and tests that try to cross tenants.

Prompt injection and output safety

What the buyer is checkingWhat stops a malicious document or user input from making the AI leak data or take actions it shouldn’t.

Evidence to attachLeast-privilege tool access, output filtering, and results from adversarial testing.

Human oversight

What the buyer is checkingWhere people review or approve what the AI produces, especially before it acts.

Evidence to attachThe approval steps in the product, and who is accountable for AI features.

Logging, monitoring and incidents

What the buyer is checkingWhat is logged, how long it is kept, who can see it, and how an AI incident is handled.

Evidence to attachRetention settings, access controls on logs, and an incident process that names AI cases.

Governance and policy

What the buyer is checkingThat someone owns AI risk, and that a policy covers how AI features are approved and changed.

Evidence to attachAn AI policy that matches the systems, a named owner, and a record of reviews.

The questionnaires we work with.

Whatever the format, we answer from your systems and keep the answers consistent across buyers.

Shared Assessments SIG

A widely used third-party risk questionnaire. Recent versions include questions on AI.

CSA CAIQ and AI-CAIQ

The Cloud Security Alliance’s questionnaires. The AI-CAIQ goes with its AI Controls Matrix.

Custom buyer forms

Spreadsheets and vendor-risk portals, often with a few AI questions added by the buyer’s AI or privacy team.

How it runs.

We start from a real questionnaire with a real deadline, so the work is tied to a deal.

  1. Step 1

    Send the questionnaire

    The live one you’re stuck on, with the buyer’s deadline.

  2. Step 2

    Check each answer

    We look at the systems behind every AI question and mark what you can claim today.

  3. Step 3

    Answer and fix

    You get accurate answers. Where the honest answer is weak, our engineers fix the gap.

  4. Step 4

    Reuse it

    Answers and evidence go into a library your team can use for the next buyer.

AI questionnaire FAQs.

Which questionnaires can you help with?

The AI sections of standard questionnaires such as the Shared Assessments SIG and the Cloud Security Alliance's AI-CAIQ, and the custom spreadsheets and portals that enterprise buyers send. The questions differ in wording but come back to the same topics: data use, model providers, isolation, oversight, logging and incident response.

What if the honest answer to a question is no?

Say so, and say what you are doing about it and by when. Buyers accept a clear no with a plan far more often than a vague yes they can’t verify. Where the fix is engineering work, we can do it so the next answer is yes.

Do you need access to our systems?

To answer accurately, yes, read-only and limited to what the questions cover. We check the settings, logs and data flows behind each answer instead of relying on what people remember. Access is agreed up front and removed when we finish.

Do you certify us for ISO 42001 or any other standard?

No. We are not a certification body. We help you put the controls and evidence in place, and we can prepare you for an audit by an accredited certification body if you decide to get certified.

Is this legal advice?

No. We work on the engineering and the evidence: systems, settings, logs and documents. Contract terms and legal questions go to your counsel, or to our partner law firm if you want us to involve them.

Send us the questionnaire that’s holding up a deal.