Is it safe to upload documents to online PDF and file tools?
The honest answer depends on one question most tools never make obvious: does your file actually leave your device — and if it does, where does it go and for how long?
What "upload" actually means
When you drop a file into most online PDF or image tools, it travels across the internet to the company's servers, gets processed there, and a result is sent back. For a moment — and sometimes much longer — a copy of your document lives on a machine you don't control, in a country you may not know, governed by a privacy policy you probably didn't read. That is the real meaning of "upload": not just a progress bar, but a transfer of custody. A small but growing class of tools works differently — they run entirely in your browser using JavaScript and WebAssembly, so the file never leaves your device at all. Whether a tool is server-based or browser-based is the single most important thing to establish before you trust it with anything sensitive.
The two real risks: retention and training
Even reputable upload-based services keep your file temporarily — iLovePDF, for example, states it deletes uploaded files within about two hours. That's a deliberate retention window, not zero retention, and policies vary wildly: some free tools keep files far longer than necessary, occasionally indefinitely, which turns them into a tempting target for attackers. The second risk is training. Consumer AI tools like ChatGPT may use your uploaded content to improve their models unless you opt out, while enterprise tiers are excluded by default — so the same product can treat your file very differently depending on your plan. For a document under an NDA or covered by HIPAA, "deleted in two hours" and "maybe used for training" are both compliance questions you have to be able to answer.
How to vet an online tool before you upload
Open the privacy policy and find three specific things: the retention window (how long files are stored and whether deletion is automatic), the training clause (whether your content trains any model, and whether you're opted in or out by default), and the jurisdiction (which country's laws and which subprocessors apply). If any of the three is vague or missing, treat that as a no for sensitive material. A faster, more honest test exists for browser-only tools: turn on airplane mode and try the tool. If it still works fully offline, the file physically cannot have been uploaded — there's no connection to upload it over. That's a property you can verify yourself, not a promise you have to take on faith.
The redaction trap most tools fall into
This is where uploading is not even the worst danger — the tool itself can betray you. Drawing a black box over text in a typical PDF editor only adds an annotation on top of the rendering; the original words remain in the file's text layer, fully selectable, searchable, and copyable. Anyone can select the area under the box, copy, and paste it into Word to reveal what was hidden. It is not theoretical: Paul Manafort's lawyers exposed sealed text this way in 2019, and the 2025 Epstein file releases showed the same failure again, with "redacted" passages recovered by simple copy-paste. True redaction must remove the underlying text from the document's content stream entirely, so there is nothing left to extract by copy, search, OCR, or code.
And the hidden data you forget about
Redaction failures are loud; metadata leaks are silent. PDFs carry author fields with full legal names, system usernames, employer details, file paths, and sometimes tracked changes or revision history that reveal what a contract said before it was softened. Government bodies have published "redacted" PDFs whose metadata still exposed names and locations to anyone with basic forensic tools. So safe document handling is really three habits stacked together: don't hand custody of the file to a server you can't vouch for, redact in a way that destroys the underlying text rather than covering it, and strip metadata before you share. Each step is only as private as the place it runs.
The on-device alternative
The cleanest way to remove the upload risk entirely is to never upload. nimbril is a suite of tools that run completely in your browser — files are never sent, never retained, never used for training. We're a newcomer, not a household name, so we'd rather you verify than trust: switch to airplane mode and the tools still work. For the highest-stakes step, our Redact tool flattens your redactions into the pixels and exports an image-only PDF, so the covered text is gone from the file's structure, not just hidden behind a box that can be deleted. That makes the result truly irreversible — and because it happens on your machine, the document never leaves it in the first place.
Frequently asked
Is it ever safe to upload a confidential document to an online tool?
Yes, conditionally. If you've checked the retention window, confirmed your content won't be used for training, and you're comfortable with the vendor's jurisdiction, an ephemeral upload tool can be acceptable. But for documents under an NDA, HIPAA, or any compliance obligation, the default-safe choice is a browser-only tool where the file never leaves your device — there's nothing to retain, breach, or train on.
Do online PDF tools delete my files after processing?
Reputable ones delete on a published schedule — iLovePDF, for instance, removes uploads within roughly two hours. But policies vary enormously, and some free tools keep files far longer or indefinitely. "Deleted soon" still means your file existed on someone else's server. Browser-only tools sidestep the question entirely because the file is never uploaded.
Will my uploaded document be used to train AI?
It depends on the tool and your plan. Consumer AI services such as ChatGPT may train on your uploads unless you opt out, while their enterprise tiers exclude customer content by default. Dedicated file tools usually don't train on documents, but you should confirm it in the privacy policy. On-device tools like nimbril can't train on your files because the files never reach a server.
Why isn't a black box over text real redaction?
In most PDF editors a black box is just an annotation layered on top of the text. The original words stay in the file's text layer, so anyone can select under the box, copy, and paste to reveal them — exactly how sealed text leaked in the Manafort filing and the 2025 Epstein releases. Real redaction removes the text from the document's content stream so nothing can be recovered.
How can I verify a tool really runs on my device?
Turn on airplane mode and use it. If every feature still works with no internet connection, the file cannot be uploaded — there's no link to send it over. This is a test you can run yourself, which is far stronger than trusting a marketing claim. nimbril's tools are built to pass it.