Get It In Writing

Is It Safe to Upload a Contract to ChatGPT? The Privacy Risk Nobody Mentions

Everyone’s doing it now. You get a contract, you don’t fully understand it, so you paste it into ChatGPT and ask what you should be worried about. It feels smart and harmless — you’re just trying to protect yourself.

Here’s the part nobody mentions: the moment you upload a document, you’ve sent it to servers you don’t control, and you may have just broken a promise you made to someone else. As a business attorney, that second problem is the one that keeps costing my clients, and almost nobody sees it coming.

Two different risks, and you need to think about both

Risk 1: Where the document goes. Whatever you type or upload is sent to servers owned by the company that runs the tool. It can be exposed in a data breach or produced in response to a legal request. And unless you’ve specifically opted out, what you share can be used to train the model — meaning your content gets absorbed into the system itself. I was reading about how providers of cloud services are feeding their clients’ content into large language models: if you save documents to a cloud drive, the provider may hand that content to their model. And here’s the ugly part — even if you turn the feature off later, the damage may already be done. The model was already trained. You can’t un-ring that bell.

Risk 2: You may not have the right to share it in the first place. This is the one that turns a privacy issue into your legal problem. Lots of the documents people paste into AI are exactly the ones they promised to keep private:

  • A contract with a confidentiality clause — you agreed not to disclose it, and uploading it to a third-party service is disclosing it.
  • Something covered by an NDA you signed.
  • Documents your employer’s policy forbids putting into outside AI tools.

So in the act of “protecting yourself,” you can breach a confidentiality obligation, violate an NDA, or break a workplace policy. The tool doesn’t warn you. It just says “Sure!” and gets to work.

A real example of why context matters

Someone I know uploaded a confidential agreement that was one piece of a much larger transaction. ChatGPT essentially yelled at her: “this is a major legal risk for you and you should not sign it as-is without a lawyer reviewing it.”

Sounds helpful, right? Except the AI had no idea it was looking at one document out of many. It didn’t have the other agreements that made up the deal, or the context that explained why that clause was there. So it produced alarm without understanding — and she’d also just put a confidential document she may not have had the right to share into a third-party system. She got the worst of both worlds: a scary, contextless answer and a potential breach.

“But people accept what the AI says anyway”

That’s the other half of the trap. People take what AI tells them and act on it without verifying — not realizing it’s working from partial information and can be confidently wrong. If it’s something that really matters, you have to spend the time to verify it. Ask it to cite sources you can check, or feed it only material you’re actually allowed to share and tell it to answer solely from that. (For what AI gets wrong when it drafts your agreements, see can I use ChatGPT to write a business contract?)

How to use AI on contracts without shooting yourself in the foot

  • Check for confidentiality restrictions first. Does the document, an NDA, or your job’s policy prohibit sharing it? If yes, don’t upload it. (When you actually need an NDA — and what it should say.)
  • Strip identifying details if you only need general help understanding a clause — remove names, numbers, and specifics.
  • Turn off training / use privacy settings where offered — but treat anything you upload as potentially permanent, because it may be.
  • Never assume the answer is complete. The AI only knows what you gave it. Deals are made of many documents and a lot of context it can’t see.
  • For anything that actually matters, use a human who can see the whole deal and is bound to keep it confidential.

The bottom line

Is it safe to upload a contract to ChatGPT? Technically it usually works. But “it works” and “it’s safe” aren’t the same thing. You’re handing a copy to a company’s servers — possibly forever — and you may be breaking a confidentiality promise while you do it. For a document that doesn’t matter, fine. For a real business agreement, the convenience isn’t worth trading away your privacy and your own contractual obligations.

The safest documents are the ones that were built right in the first place — so you don’t need a chatbot to tell you what you should have been worried about.

Want to know where your business is actually exposed? Grab the free Get-It-In-Writing Business Audit — 15 things that should be in writing, but usually aren’t. [Get the checklist →]

This article is general education, not legal advice, and does not create an attorney-client relationship. Laws vary by state and situation — for your circumstances, talk to a licensed attorney.