Can you upload customer drawings and specifications to an AI?
Drawings and specifications from the customer are confidential and are often covered by a non-disclosure agreement. What to check before uploading them to an AI: the NDA, data processing, where processing takes place and training.
Customer drawings and specifications do not belong in a private AI account. In many cases they may go into a system the company has approved, if three things hold: the NDA with the customer allows service providers, there is a data processing agreement with the AI vendor, and the contract guarantees processing in the EU and excludes training.
Contents
A design engineer receives a package from the customer: the specification, a layout and twelve part drawings of the existing plant. He wants to know whether the transfer height in the layout matches the drawings. An AI promises the quickest answer. The question that often gets skipped is whether the package may go there at all.
This article is not legal advice. It sorts the questions you should settle before customer documents end up in an AI and shows who can answer them.
Why are customer drawings a special case?
Your own documents belong to you, and you decide what happens to them. Customer documents belong to the customer. You received them for a specific purpose, usually for the quotation and the order, and often under a non-disclosure agreement.
On top of that, drawings and specifications rarely contain only technical content. The title block holds the names of the people who drew and checked the drawing, the specification lists contact persons with phone numbers. That means personal data is involved, and the GDPR applies.
What does the non-disclosure agreement say?
This is the first and most important question. Many NDAs in machine and plant engineering allow disclosure to employees and to service providers who are themselves bound to confidentiality. Some explicitly exclude cloud services or require consent.
Read the agreement with three points in mind:
- Who may see the documents, only your own employees or service providers as well?
- Are there rules on where the documents may be processed or stored?
- Does the customer have to consent before the documents are passed on?
If the agreement is unclear, ask the customer. Many customers now have their own rule on the use of AI, and a short question costs less than a loss of trust.
What has to be agreed with the AI vendor?
Data processing agreement. As soon as personal data is processed, you need an agreement under Art. 28 GDPR. It sets out what the vendor may do with the data and which subprocessors are involved.
Place of processing. Where are the documents stored, and where does the language model run? Both should be in the EU. Whether the operator belongs to a US group is a separate question for your data protection officer to assess.
Training. Using the documents to train models must be excluded by contract, not just by a setting.
Deletion. What is deleted when the project ends, and how quickly? Document copies, the search index and caches all count.
A checklist with these points is on the page about GDPR-compliant AI.
Why a private account is never the answer
A private ChatGPT account belongs to the person, not to the company. There is no contract between your company and the vendor, no data processing agreement and no central control over what gets uploaded. None of the questions above can be answered with a yes.
The real risk arises when there is no approved route. Then design engineers use their private one, because the work is pressing. More on this in the GDPR check on ChatGPT for internal documents. The article on GDPR-compliant ChatGPT alternatives gives an overview of routes that can be approved. Which tasks an approved system can take on in design is shown in the guide AI in engineering design.
What to settle internally
- Name one approved tool in which customer documents may be processed.
- Decide which documents may go in, for example anything under an NDA only after the agreement has been checked.
- Inform the customer if the agreement requires it or is unclear.
- Have projects deleted after completion if the customer requires it.
How does KoAssist handle customer documents?
KoAssist, the AI agent for machine and plant builders, is built for working with specifications, drawings and project folders. The documents are held in ISO 27001 certified data centres in Germany, and language processing runs exclusively on EU infrastructure. Training on customer data is excluded by contract, and the data processing agreement under Art. 28 GDPR is standard. On written request, data is deleted within 48 hours, with a deletion protocol. Permissions come from your file storage. Anyone who may not open a document there will not see it in KoAssist either.
Data flow and subprocessors are described on the page about security and EU hosting. For discussions with your customer or your IT department, we provide the security document. If you would like to see KoAssist on one of your own projects, you can book a demo.
Frequently asked questions
More from the blog
Back to overview- 15 November 2026Article5 min
Can ChatGPT analyse a customer specification?
An 80-page customer specification, and the quotation has to go out by Friday. What ChatGPT does well in the analysis, where it reaches its limits and what to settle before a customer specification ends up in a chat window.
5 min - 15 November 2026Article4 min
Backing change orders in plant engineering: how additional work from customer changes gets paid
Customer changes create additional work, and it only gets paid when it is backed by evidence. What a change order in plant engineering needs, when to submit it and which documents make the difference at the final invoice.
4 min - 15 November 2026Article4 min
How much time does coordination cost in design engineering?
Collecting status, answering questions, preparing status meetings: in design departments, coordination takes up many hours a week. Two cases from special-purpose machine building and a simple method for measuring your own effort.
4 min
