AI questions in customer audits and supplier questionnaires: answers and evidence

What customers ask about AI, and why

Customers now ask their suppliers which AI tools process the customer's data, where those tools run, whether inputs are used to train a model, how long prompts and answers are kept, who can read them and which subcontractors are involved. The questions arrive as a new section in the supplier questionnaire, as an AI clause in a contract or as a topic in a customer audit. A supplier that runs its AI on its own server can answer most of them from its own records: where the server stands, who has access, what the query log contains and when it is deleted.

This guide is for owners, managing directors, quality managers and sales managers at suppliers that receive these questionnaires: machine builders, component makers, engineering offices, contract manufacturers. It explains the questions in plain words, the documents that answer them, and where an in-house AI server changes the answer. The server itself comes in one section near the end. Legal points are general information on EU law as of October 2026; your legal counsel or data protection officer confirms what your contracts and your situation require.

Three developments push customers to ask. Companies in scope of the NIS2 Directive must take "supply chain security" measures, which Article 21(2)(d) defines as including the relationships with "direct suppliers or service providers", and they pass that duty on as questionnaires, as our guide to NIS2 supply chain security describes. Where a supplier processes personal data on a customer's behalf, the GDPR applies: the supplier is then a processor under Article 28 and may engage another processor, such as a cloud AI service, only with the customer's prior authorisation. And many customer contracts and NDAs limit to whom confidential drawings, specifications and prices may be passed. An AI tool that sends documents to an outside service touches all three.

AI questions in information security questionnaires, in plain words

Questionnaires word their AI questions differently, but most come down to the same eight. The table gives each one in plain words, what the customer wants to learn and the record that answers it. A query log is the record of who asked the AI what and when, with the answer; training on inputs means a provider uses your prompts to improve its model; a subprocessor is a company that processes the customer's data on your behalf.

Our summary of typical AI questions; wording varies by customer and questionnaire.

How detailed a questionnaire can become shows in the Cloud Security Alliance's AI Controls Matrix. Version 1.1, presented in a CSA blog post of 14 July 2026, has 247 control objectives across 18 domains and comes with a questionnaire, the AI-CAIQ, of 320 questions for self-assessment and third-party assessment. It sorts its guidance by role, from model provider to AI customer. A supplier will rarely receive the whole set, but a customer may draw its own questions from a framework of this kind.

The question on public chatbots needs care. Staff who paste a customer's drawing list into a chatbot on a personal account do so under consumer terms that the company has not reviewed, as our guide to shadow AI and a usage policy explains. A truthful answer needs the policy, the approved tool and the training record behind it.

Where a cloud AI service and an in-house server differ in the answers

With a cloud AI service, several answers depend on the provider's documents. OpenAI's enterprise privacy page, updated on 8 January 2026, says for its business products that "By default, we do not use your business data for training our models", that workspace administrators of ChatGPT Enterprise and ChatGPT Business control how long their data is retained, that ChatGPT Business, Enterprise and the API Platform have completed SOC 2 Type 2 audits, and that OpenAI can execute a data processing addendum with customers. A supplier can quote those statements, attach the provider's reports and name the provider as a subprocessor. Whether a given customer accepts that is a question of its contract with you.

With an in-house AI server, the supplier answers from its own records, and the customer's data stays with the supplier.

OpenAI enterprise privacy page (updated 8 January 2026, read 25 October 2026) for the cloud column; the in-house column is our summary of what the supplier controls.

An in-house server does not answer everything by itself. The customer will still ask how it is protected: sign-in through the company directory, no access from the internet, few administrators, updates and backups. Those are the questions in our guide to in-house AI security questions for management, and the answers belong in the same folder. If a service company maintains the server with remote access, it is a subcontractor too and goes on the list.

We supply AI servers built to order and set up the models under our Private AI/ML service; send us the AI section of a questionnaire you received, and our team will tell you which records an in-house server lets you produce.

Records to keep before the next questionnaire

Six documents answer most AI questions, and each should carry a date and an owner.

1. AI tool register. One line per tool: name, purpose, departments, data classes allowed (public, internal, customer-confidential, personal data), where it runs, account type, owner. It includes AI features switched on inside software you already use.
2. Data-flow sketch. One page per AI use: where the documents come from, where the model runs, where the answers go and what is logged. A customer's auditor understands a drawing faster than a paragraph.
3. Access list and review. Who can use the assistant, who can read the query log, who administers the server. The list comes from your company directory; the review records the date, the reviewer and the changes.
4. Retention setting. How long prompts, answers and uploaded files are kept, as a screenshot or configuration export, and who decided it with the data protection officer.
5. Supplier AI policy and training record. Which data may go into which tool, and who completed the training. Article 4 of the AI Act, applicable since 2 February 2025 and amended by Regulation (EU) 2026/1744, asks providers and deployers to take measures to support the AI literacy of their staff, and the Commission's Q&A says no certificate is needed and an internal record of trainings can be kept, as our guide to AI literacy training under Article 4 sets out.
6. Subcontractor list. Hardware supplier, service partner, data centre, each with its role and whether it can reach the data.

With these six on file, answering the AI section becomes a matter of copying from current documents. Keep the answers you send in a library of approved answers, because the next customer will ask the same questions in other words. Our guide to AI governance with ISO/IEC 42001 shows how the same records grow into a management system if a customer asks for one.

AI clauses in supplier contracts and customer data

AI clauses in supplier contracts vary. Some ban all AI tools for the customer's documents, some ban only external services, some require notice before a new tool or subprocessor is used, and some require deletion of prompts and outputs at the end of the project. Read the clause for those four points and answer with what you do. A supplier that signs a general ban and then uses a cloud tool has a contract problem, whatever the provider's terms say.

Where you process the customer's personal data, the data processing agreement governs. Article 28(3) of the GDPR requires that agreement to cover security measures, the conditions for engaging other processors and the information and audits the customer may ask for, and Article 28(4) passes the same obligations down to any subprocessor, as our guide to GDPR Article 28 and subprocessors explains. An AI tool that processes such data is a processing step that the agreement must allow. Our article on trade secrets, NDAs and AI tools covers the confidentiality side, and local AI and the GDPR what own hosting changes. Which contract allows what is a question for your legal counsel.

The AI server behind the answers

Drafting questionnaire answers is a small task for an AI. The assistant searches your approved answers, policies and certificates, proposes an answer per question with the source document, and a person checks and sends it; the method is retrieval on company documents, explained in our guide to RAG on company data. A model of moderate size does this well, so it pays to run it on the company AI server that also carries the assistant for engineering, sales and the office. The same server then appears once in the tool register, with one location, one access list and one query log.

Users-per-card rule (0.9 × driver-visible memory, less 3 GiB per card and less the weights): Qwen3.8-27B FP8 28.75 GiB with FP8 cache, 1.14 GiB per 32K conversation; gpt-oss-120b 60.8 GiB with 16-bit cache, 0.5625 GiB per 16K conversation; power by our estimate (cards, processors and 15 per cent for the rest), confirmed by a measurement on the delivered system.

A conversation here means one request being answered at the same moment; many more staff can use the assistant, because most of them are reading or typing at any given time. The entry server has two NVIDIA RTX PRO 6000 Blackwell Server Edition cards, two AMD EPYC processors, 384 GB of DDR5 ECC and two NVMe drives, and runs from one ordinary 230 V 16 A feed in a server room. Four cards give a tensor-parallel pair, one model spread over two cards, for several hundred staff, as our page on private ChatGPT servers by company size works out. Without a server room, our Private AI/ML service runs the platform on dedicated hardware in a Tier-3 data centre in Lithuania, inside the EU, and that location goes into the register. A company without an IT department will find the division of tasks in our guide to an AI server without an IT department.

What we supply, and what we answer about ourselves

You can buy this system from us: our team sizes the configuration to your workload and budget and replies within one business day. We build AI servers to order with two, four or eight cards, assembled and tested before delivery, delivered anywhere in the EU on one EU contract and invoice, with manufacturer warranty on every component handled through us; quantities and lead time are confirmed in writing with the quote. Our Private AI/ML service, with our engineering partner Vixen.UNO, sets up the models, the search on your documents, access control and the query log. If we work in your environment, we become a subcontractor in your own answers: our Private AI/ML page states an NDA before technical detail, a data processing agreement on request and subprocessors named in the contract. Describe the questionnaires your customers send, and we propose how to set up the server and the records together.

FAQ