Is It Safe to Use AI With Client Drawings? What to Check First

Last updated: August 2026

Putting a client’s drawings into an AI tool is not the same as using AI on your own work. It is someone else’s confidential information, usually covered by an NDA and by your professional duty to keep it private, and the free consumer version of most assistants can use what you paste to train the model. So the honest answer is: it can be safe, but only on the right tier, with the right controls, and with the client’s agreement. On a free account by default, it is a confidentiality risk.

This guide covers whose risk it actually is, what can go wrong, how the consumer and business tiers really differ, and a checklist to run before you upload anything. It sits under our roundup of the best AI tools for engineers and is the client-confidentiality companion to our pillar on AI tool data security and engineering IP.

Whose risk is it? The duty you already signed

Before any tool question, there is a duty question. If the drawings belong to a client, you are almost certainly bound by an NDA and by professional ethics to keep them confidential. The NSPE Code of Ethics is explicit: engineers must not disclose, without consent, “confidential information” concerning the business affairs or technical processes of a client. Pasting those drawings into a tool that may store or train on them is a form of disclosure, and most NDAs written before generative AI never contemplated it. That is why the safe-use question is really a consent-and-contract question first, and a settings question second.

What can go wrong

The concrete risks are specific. On a consumer tier, your inputs can be used to train the model, which means a client’s drawing could influence outputs you never see. Even when training is off, data may be retained for a period, and it can pass through third-party sub-processors. And a real-world reminder of the stakes: Samsung banned employee use of ChatGPT and similar chatbots across company devices after engineers pasted proprietary source code and internal notes into ChatGPT in three incidents within about twenty days (Forbes, 2023). That was internal code; a client’s confidential drawings on a consumer account carry the same exposure plus a contractual breach.

Consumer versus business: what actually changes

The single most important distinction is the tier, because the default behavior differs sharply. Business and enterprise tiers are built not to train on your data; consumer tiers increasingly may, unless you opt out.

The specifics, from each vendor’s own pages: OpenAI states that data sent to its API is not used to train its models unless you opt in, and it offers a zero-data-retention option on approval, with abuse-monitoring logs otherwise kept up to thirty days (OpenAI). OpenAI similarly states it does not train on ChatGPT Business, Enterprise, or API data by default (per OpenAI’s business and enterprise privacy pages). Anthropic says that by default it does not use inputs or outputs from its commercial products, such as Claude for Work and the API, to train its models (Anthropic). Its consumer tiers are different: since the August 28, 2025 consumer-terms update, chats on Claude Free, Pro, and Max can be used for training unless the user declines, with retention extending to five years when training is allowed (Anthropic). Microsoft states that Microsoft 365 Copilot prompts, responses, and Graph data are not used to train the foundation models (Microsoft).

What actually makes it safe

Safe use of a client’s drawings comes down to a few controls, layered.

• A no-training business or enterprise tier. This is the floor. It moves you from “may train unless I remember to opt out” to “contractually does not train.”
• A signed data processing agreement. A DPA is a binding contract setting each party’s obligations for protecting data, and under GDPR a processor cannot bring in a sub-processor without authorization and a further binding agreement (GDPR.eu). Review who the sub-processors are, because major assistants can appear inside other products as sub-processors.
• Zero data retention where offered. For API use, ZDR stops inputs and outputs being stored at all, which is the strongest posture for confidential material.
• Minimize what you share. Strip client names, project identifiers, and anything not needed for the task, so even a mishandled input reveals less.
• Client consent on record. Because the duty is to the client, get their agreement, and update NDAs to address AI tools explicitly.

Even “we delete it” has limits

One caveat worth knowing: a deletion or retention promise can be overridden by law. In the New York Times litigation against OpenAI, a court ordered OpenAI in 2025 to preserve output logs that would normally have been deleted, before the going-forward preservation obligation was later released (Bloomberg Law). Per OpenAI’s account, that order targeted consumer ChatGPT and the API without zero data retention, not its enterprise or ZDR offerings, which is another reason the enterprise and ZDR posture matters for confidential client work. Treat “it gets deleted” as a policy that can bend under a legal hold, not an absolute guarantee.

A watch on the topic

Do this before you upload

• Check the NDA and get client consent. Confirm you are allowed to use an AI tool on this material, and record the agreement.
• Use a no-training business tier with a DPA. Not a personal free account.
• Turn training off on any consumer tool you cannot avoid, and prefer zero data retention for the API.
• Strip client-identifying detail from what you paste wherever the task allows.
• Log what you did. Which tool, which tier, what was shared, and the consent, so the decision is defensible.

For the mechanics of whether a model trains on what you upload, see does ChatGPT train on my drawings, and for the separate question of liability on stamped work, see can I use AI for stamped drawings.

Frequently asked questions

Is ChatGPT Enterprise safe for a client’s drawings?

It is far safer than a consumer account, because OpenAI states it does not train on Business, Enterprise, or API data by default, and enterprise use can be backed by a data processing agreement and zero data retention. That still does not remove your duty: you need the client’s consent and an NDA that permits AI use, and you should minimize what you share. The tier reduces the tool risk; the professional obligation to the client remains yours.

Does the API train on the files I send?

Per OpenAI, data sent to the API is not used to train its models unless you explicitly opt in, and abuse-monitoring logs are otherwise kept up to thirty days, with a zero-data-retention option available on approval. Anthropic likewise states it does not train on commercial or API inputs by default. Confirm the current terms and your specific configuration before relying on this for confidential client material, because policies and defaults change.

What if I only use the free version once?

A single upload to a consumer tier still counts as disclosing confidential information to a third party, and on a default consumer account it may be retained or used for training unless you have turned that off. If the drawings are a client’s, one paste can breach an NDA regardless of how briefly you used the tool. Use a no-training business tier, or do not upload the material at all.

Do I need to tell the client?

Yes. Because the confidentiality duty runs to the client, the safe course is to get their consent before putting their drawings into any AI tool, and to make sure your NDA covers AI use explicitly. Many older NDAs predate these tools and say nothing about them, so treat silence as a reason to ask, not as permission.

Sources

• OpenAI, API data controls (training and retention)
• OpenAI, enterprise privacy (business and enterprise data use)
• Anthropic, is my data used for model training (commercial)
• Anthropic, updates to consumer terms (August 2025)
• Microsoft, data, privacy, and security for Microsoft 365 Copilot
• GDPR.eu, what is a data processing agreement
• NSPE Code of Ethics (confidentiality, III.4)
• Forbes, Samsung bans ChatGPT after code leak (2023)
• Bloomberg Law, OpenAI ordered to preserve ChatGPT logs

Written by the CognitiveFuture editorial team. We build our guidance from vendors’ own policy pages and published sources, and we label vendor commitments as such. We do not independently audit any vendor’s data handling, and vendor terms and defaults change, so confirm the current policy and your own configuration before using an AI tool with confidential material. This is general information, not legal advice; your NDA and professional obligations to the client govern.

About the author

Richard Johnson

Richard Johnson is an AI specialist at one of the world's largest technology companies, where he has spent the past three years helping organizations adopt AI. CognitiveFuture extends that work publicly: gathering the available evidence on each tool, from vendor documentation to independent reviews and user feedback, and cutting a crowded market down to the right choice for the job in front of you.