Working with AI

Can you upload confidential documents to ChatGPT?

A practical look at sensitive files, and a safer way to work

If you work with contracts, filings, financial reports, or medical records, you've probably wondered whether it's fine to drop one into ChatGPT and ask it to summarize. The interface makes it feel harmless. The consequences aren't always.

What actually happens when you upload a file

When you upload a document to a hosted AI tool, the file is sent to that company's servers to be processed. That's not a criticism of any one provider — it's simply how a cloud service works. But it means a few things are now outside your direct control:

• The file is transmitted and stored on infrastructure you don't own, at least for as long as it takes to process — and often longer, depending on the service.
• Whether your content can be used to improve the model depends on the specific product, plan, and account settings. Consumer plans, team or enterprise plans, and the API can all have different terms.
• Those terms change over time. What was true last year may not be true today.

Because these details vary and shift, the only reliable source is the provider's own current documentation. If you're going to upload sensitive material, read OpenAI's data-usage and retention pages for the exact plan you're on before you do — not a blog post, not a forum thread.

Why this is different for regulated work

For most people, uploading a PDF is a reasonable convenience. For some roles, it can be a problem regardless of any single provider's policy:

• Lawyers and legal staff owe a duty of confidentiality to clients. Sending privileged filings or contracts to a third-party service can conflict with that duty and with firm or court rules.
• Finance professionals handling non-public information — draft filings, deal documents, internal figures — operate under rules about where that information can go.
• Healthcare workers deal with patient records that are protected by law in most jurisdictions.

In these cases the question isn't "is ChatGPT trustworthy?" It's "am I allowed to send this document to any outside service at all?" Often the answer is no, or only under specific approved arrangements — and a personal ChatGPT account usually isn't one of them.

The safer pattern: keep the file local, share only what's needed

Here's the thing most people miss: you rarely need an AI model to see the whole document. You need help with a few specific pages — a clause, a section, a table. The safer habit is to do the file handling on your own device, and only ever put in front of an AI the minimum you're actually permitted to share.

That splits into two separate decisions:

• Preparing the file — opening it, reading it, pulling out the relevant pages. There's no reason this step needs to touch anyone's servers. It can happen entirely on your own computer.
• Sending content to an AI — a deliberate, separate choice you make only for material you're cleared to share, on a service and plan whose terms you've actually checked.

Keeping those two steps apart is what gives you control. The preparation never has to be a data-sharing event at all.

How to prepare a PDF without uploading it anywhere

This is exactly what PDF Source Prep is built for. It runs entirely in your browser — your file is opened and processed on your own device and is never uploaded to any server. You can open a long PDF, pick out just the pages you need, and export a clean packet, all locally. Then you decide, separately and deliberately, what (if anything) is appropriate to hand to ChatGPT or Claude.

It won't make a decision about confidentiality for you — that's yours to make. But it removes the step where the whole document gets sent somewhere just so you can trim it down.