AI Vendor Security Questionnaire Generator
Export a 12-point AI-specific vendor evaluation questionnaire in clean Markdown format to send to vendor security teams during procurement.
Customization Options
Included Audit Sections
How to use this questionnaire
Send this questionnaire to vendor procurement or security contacts. Once completed, compare vendor responses against our guide on how to judge an AI provider and our published AI provider dossiers.
Why a general security questionnaire is not enough
Standard vendor questionnaires were written for software that stores and transmits data. They ask about encryption, access control, certifications and breach notification — all necessary, and none of it reaching the questions that actually distinguish one AI vendor from another.
The AI-specific risks live in different fields: whether your inputs train the model, how long prompts are retained and in which store, whether an abuse-monitoring carve-out survives a zero-retention commitment, which downstream model provider ultimately receives your data, and whether any of it is contractual or merely described on a marketing page. A SOC 2 report answers none of those.
Time it for procurement, not for security review
The single most common mistake is running this assessment after the organisation has standardised on a tool. At that point the security function's options are to accept the terms or fight a battle it will lose — because with most large providers on standard business plans, the DPA is a take-it-or-leave-it document and no amount of redlining changes it. Your leverage is the choice of provider and tier, and that choice is made at procurement.
Reading the answers
Precision matters more than the answer itself. A vendor that responds with a specific retention period, a named sub-processor list and a documented export mechanism is one you can hold to something. A vendor answering "industry-standard practices" and "as long as necessary" has told you it has either not done the work internally or does not want to be held to it.
Watch for three gaps in particular. Scope — a certification covering the corporate IT environment is not one covering the production platform processing your data. Tier — a commitment that applies to enterprise plans may not apply to the seats your team actually uses. Carve-outs — "we do not train on your data" and "nobody at this company will ever look at your data" are different sentences, and the gap between them is usually an abuse-monitoring retention window disclosed in a different document.
Keep the completed responses with the executed DPA and the date. That package is the vendor risk assessment record an auditor expects to have existed at onboarding, and it is considerably harder to reconstruct afterwards.
Related reading
The AI vendor security questionnaire guide explains each question and what a good answer looks like. How to get a DPA for an AI tool covers the contractual step that follows, including the five retention questions that are not in any DPA. AI tools in a SOC 2 audit covers the evidence this record becomes.