Uploading a privileged document to an online redaction tool feels efficient. It is also the moment a copy of your client’s most confidential data lands on someone else’s servers — and for most legal work, that alone is the breach. Here is how to redact PII from legal PDFs properly, without your documents ever leaving your machine.
Key facts
- Uploading a document to an online redaction tool sends a copy of its contents to a third-party server — which, for privileged legal material, can itself be a confidentiality breach.
- True redaction removes the underlying text; a black box drawn in a generic PDF editor usually leaves the original text extractable by copy-paste.
- On-device redaction — desktop or client-side in the browser — keeps the document on your machine while still automating detection of names, IDs, and other PII.
- Reversible tokenization lets you share a document with identities removed and decode it back later, with the mapping stored locally.
The short answer
To redact personal data from legal PDFs without sending them to a third party, you need a tool that does two things: it runs detection and redaction on your own device — desktop or fully client-side in the browser — and it removes the underlying text rather than drawing a black box over it. Uploading privileged documents to a cloud service means a copy now lives on infrastructure you don’t control, which is a confidentiality and GDPR problem regardless of the vendor’s promises.
Why “just use a cloud tool” is the wrong instinct for legal work
Most online PDF redaction tools upload your file to their servers, process it, and hand back a download. That is fine for a restaurant menu. It is a problem for a deposition transcript, a settlement agreement, or a client’s medical records:
- Confidentiality and privilege. The moment the file leaves your device, you have disclosed its contents to a third party. Depending on your jurisdiction and bar rules, that can breach your duty of confidentiality — even if nothing is ever leaked.
- GDPR / data-protection exposure. Uploading documents containing personal data to a processor makes that vendor your data processor, requiring an agreement, transfer safeguards, and a lawful basis. Most firms never paper this for an ad-hoc tool. (More on what actually counts as compliant in our guide to GDPR document anonymization.)
- You can’t un-ring the bell. Even with deletion policies, you are trusting a stranger’s infrastructure with the most sensitive thing you handle. One misconfigured bucket is a career-ending incident.
The fix isn’t a better cloud vendor. It is not uploading the document at all.
The trap nobody tells you about: black boxes aren’t redaction
Here is the technical gotcha that catches even experienced legal teams. If you “redact” a PDF by drawing black rectangles over names in a generic PDF editor, the text underneath is usually still there. Anyone can copy-paste it out, or extract it with a one-line script.
This is not hypothetical — it has happened in high-profile filings repeatedly:
- In January 2019, Paul Manafort’s defense team filed a court document whose redactions could be defeated simply by copy-pasting the blacked-out passages into another program — inadvertently revealing sensitive details the redactions were meant to hide. The story made national headlines.
- The TSA once published its airport screening manual online with “redacted” sections that were merely black-highlighted over still-selectable text, exposing sensitive procedures to anyone who copied them out.
The lesson is the same every time: a visual mark is not redaction. True redaction requires destroying the underlying content, not hiding it:
- Locating the exact text (or image regions, for scanned pages) that contain personal data.
- Removing that content from the document’s text and object layers.
- Rendering a replacement — a block, a label, or a token — in its place.
Any redaction workflow that skips step 2 is theater.
How to redact locally, step by step
- Keep the file on your machine. Use a desktop tool, or a browser-based tool that processes the document client-side (in your browser’s memory) and never transmits it. If it works with your Wi-Fi off, it is genuinely local.
- Detect the personal data. Manual find-and-highlight works for a two-page letter and collapses on a 300-page disclosure bundle. Automated detection catches names, addresses, national IDs, dates of birth, account numbers, emails, and phone numbers — combining pattern rules for structured data with language models that recognize names and organizations in context.
- Review before you commit. Automated detection should propose, not silently decide. A human check on the flagged items is the difference between defensible redaction and an accidental disclosure.
- Apply true redaction. Remove the underlying content and render the replacement. Then verify by trying to copy text out of the “redacted” region — if nothing comes out, it is real.
For scanned documents this pipeline needs an extra step; see how to redact scanned PDFs and images.
Reversible redaction: the feature legal teams actually need
Plain redaction is one-way. But legal work often needs something subtler: share a document externally with names removed, while keeping an internal, reversible mapping so your own team still knows who “[PERSON_14]” is. This is reversible tokenization, and it lets you:
- Send anonymized documents to opposing counsel, experts, or vendors.
- Keep a consistent token for the same person across an entire bundle — so “[PERSON_14]” is the same individual on page 3 and page 280.
- Reverse it internally when you need the real identity back.
Crucially, that mapping should live on your machine, not a vendor’s — the same principle as the redaction itself.
Cloud vs. manual vs. on-device: a quick comparison
| Capability | Cloud redaction tools | Manual PDF editor | On-device automated |
|---|---|---|---|
| File leaves your machine | Yes | No | No |
| Confidentiality / GDPR exposure | High | Low | Low |
| Removes underlying text | Varies | Often not | Yes |
| Handles long bundles | Yes | Poorly | Yes |
| Reversible / consistent tokens | Rare | No | Yes |
| Works offline | No | Yes | Yes |
Frequently asked questions
Is drawing a black box in a PDF editor safe redaction?
Usually not. Unless the tool explicitly removes the underlying text, the original content is still extractable by copy-paste or a script. Use a tool that destroys the content beneath the mark.
Does on-device redaction mean lower-quality detection?
No. Modern detection models run locally on a normal laptop. The trade-off used to be real; it isn’t anymore.
Is a browser-based tool actually private?
Only if it processes the file client-side and never uploads it. A quick test: disconnect from the internet — if the tool still redacts, it is genuinely local.
How do I redact a scanned PDF?
Scanned pages are images, so the tool must OCR them, locate the personal data as image regions, and redact those regions — then, ideally, keep the document searchable. We cover this in redacting scanned PDFs.
The bottom line
For anyone handling privileged or personal data, the right redaction tool is the one that never asks you to upload the document in the first place — and that removes content instead of hiding it. Local, on-device redaction gives you cloud-tool convenience without cloud-tool risk.
promptShield redacts PII from PDFs and Office files entirely on your own device — desktop or client-side in the browser — and runs 100% offline, with reversible tokenization so you can share anonymized documents and decode them later. Try it on your own document — no upload, no download required.
