Black Boxes That Aren't Redaction: The Malpractice Risk in Free Tools

Here's the scenario that keeps defense attorneys up at night: you "redact" a police report by drawing black boxes over a client's address in Adobe Reader, turn the PDF over in discovery, and opposing counsel extracts the hidden text in thirty seconds. The redaction failed — and now you have a malpractice problem.

The root cause is the difference between a visual overlay and true redaction. Drawing black rectangles in Adobe Reader "does not constitute redaction. The text remains in the file and can be easily extracted," according to AI-Redact's guide. A true redaction tool deletes the text, metadata, comments, hidden layers, and embedded objects from the PDF's data structure itself, so nothing can be recovered or searched.

Most of the "free" methods defenders reach for first fall into the overlay trap:

• Black rectangles drawn in Adobe Reader — Reader cannot redact at all; the capability exists only in paid Acrobat Pro.
• Mac Preview — its redaction behavior "has been inconsistent across macOS versions," and security professionals do not recommend it for sensitive documents.
• LibreOffice Draw — technically possible, but manual, tedious, and unreliable in rendering.
• Many online PDF editors — several vendor guides note they "mostly apply visual overlays rather than true redaction."

There's a simple test to run before anything leaves your office: open the redacted PDF and try to select or search for the text you blacked out. If it's selectable, searchable, or copy-pastable, your redaction failed — the data is still in the file.

The stakes here are concrete. Jeffrey Schoenberger, a senior consultant writing for The Missouri Bar, notes that proper redaction "protects your clients, complies with confidentiality rules, and guards against sanctions or malpractice claims." And the risk extends beyond the visible page: hidden metadata — including comments, form fields, and even embedded AI-generated metadata — can carry sensitive content you never intended to disclose, which is why sanitization of hidden information is a separate, often-missed step.

One more wrinkle specific to discovery work: most of the documents in a criminal case file are scans with no text layer at all. Automated search-and-redact only works on PDFs that have been OCR'd, and some free sanitization tools strip hidden text layers so thoroughly that the output is no longer searchable — trading one problem for another. This is the gap that a workflow like Evidence Copilot addresses: it OCRs scanned PDFs into searchable, citable text while keeping bounding boxes for redaction, produces every redaction as a new derived file so originals stay immutable, and audits each action — so the file you turn over is verifiably clean, not just visually black.

Run the select-and-search test on every redacted document before turnover. If the text comes back, so will the problem.

The Actually Free Options: Freemium Caps and Open-Source Tradeoffs

If your discovery turnover deadline is Friday and your budget is zero, you still have real options — but they split into three very different tiers, and picking wrong can leave client PII extractable from a PDF you already filed.

Tier 1: Freemium web tools with usage caps. These are the fastest path for a short police report or a few scanned pages. Smallpdf's free plan offers drag-and-drop redaction with no signup required and automatic file deletion after one hour. AI-Redact's free tier handles PDFs up to four pages with no signup and no watermarks, auto-detecting names, emails, phone numbers, SSNs, addresses, and financial account numbers. Justee is free to start, detecting over 30 PII types, and Xodo by Apryse offers a free plan as well. The catch is obvious: a 4-page cap doesn't help with a 60-page incident supplement bundle, and paid tiers start adding up (AI-Redact charges $0.08 per page; Justee's paid plans run from $16/mo annually).

Tier 2: Fully free open-source tools. Two options cost nothing at all, but demand technical comfort:

• Microsoft Presidio — MIT-licensed, Python-based NER plus regex detection of PII. Free forever, but it requires development setup and won't run in your browser.
• First Look Media's pdf-redact-tools — runs offline, converts PDF pages to images, and sanitizes hidden content. The tradeoff: it performs no PII detection, and its output is not text-searchable afterward.

That last caveat matters more than it sounds. Automated search-and-redact only works on PDFs with an OCR text layer — without one, the tool can't find the text to remove. Free sanitization tools that strip hidden layers for security destroy searchability entirely, so you lose the ability to keyword-search the very discovery you just redacted. That OCR gap is the reason a workflow like Evidence Copilot's exists: scanned PDFs and photos become searchable, citable text with bounding boxes preserved for redaction, and every redaction produces a new derived file while the original stays immutable and audited.

Tier 3: The "not actually free" traps. Adobe Reader cannot redact — the feature is exclusive to Acrobat Pro. Mac Preview's redaction behavior has been inconsistent across macOS versions, and security professionals do not recommend it for sensitive documents. LibreOffice Draw technically works but is manual, tedious, and renders unreliably. Drawing black rectangles in any of these leaves the underlying text intact and extractable.

One caution cuts across all three tiers: data handling varies dramatically. Some tools process documents entirely in-browser; others upload your client's discovery to cloud servers, where some store files indefinitely. Before uploading anything sensitive, verify the privacy policy and prefer tools with stated auto-deletion. And whatever tool you use, run the verification test before turnover: open the redacted output and try to select or search for the text you blacked out. If it's selectable, the redaction failed — and automated detection is best-effort in every tool, so human review of the output still matters.

The OCR Problem: Why Automated Search-and-Redact Breaks on Scanned Discovery

Here's the problem nobody warns you about until you're three hours into a 200-page scanned police report: Acrobat's find-and-redact only works when the PDF actually has a text layer. According to guidance from The Missouri Bar, the "Find text and redact" feature requires optical character recognition — meaning the PDF must already be searchable, copyable text before Acrobat can automate anything.

Most discovery isn't. Scanned police reports, faxed medical records, and photographed exhibits arrive as flat images with no text layer at all. Acrobat's OCR can transform scanned documents into editable text, but Acrobat Pro runs from $19.99 per user per month — it's not free, and it's not in the budget of every solo defender or public-defender office.

The free alternatives fail differently. First Look Media's pdf-redact-tools runs offline and converts PDF pages to images, which is genuinely secure — but the output is not text-searchable. You've traded a redaction problem for a review problem: a 200-page report you can no longer search for "Miranda," "consent," or the informant's name. For defenders preparing a motion to suppress or cross-exam, that's a real cost.

So the choice looks like this:

• Pay for Acrobat Pro and run OCR first, then automate find-and-redact on searchable text.
• Use free sanitization tools like pdf-redact-tools and lose searchability entirely.
• Redact manually, page by page, on documents where automated detection can't even see the text.

The workflow gap is real, and it's why Evidence Copilot built its OCR pipeline the way it did: scanned PDFs and photo evidence — receipts, exhibits, screenshots — become searchable, citable text, with bounding boxes preserved so redaction regions stay anchored to the original page. You get the searchability that free sanitization tools destroy, plus the automation that flat scans otherwise make impossible. Redaction produces a new derived file; the original stays immutable and every action is audited.

Two cautions apply no matter which path you take. First, automated detection is best-effort in every tool — as one review of free PII redaction tools puts it, human review of the output still matters. Second, verify the result yourself: try to select or search for the "redacted" text. If it's selectable, the redaction failed — and on discovery turnover, that failure is a malpractice exposure, not an inconvenience.

A Redaction Workflow for Discovery Turnover: Verify, Sanitize, Audit

The moment before you hand a redacted PDF to opposing counsel or file it with the court is the moment most redaction failures happen — not during the blacking-out, but during the turnover. A workflow that ends at "I drew the boxes" is a workflow that produces malpractice exposure.

Run the same verification test on every tool's output, free or paid: open the redacted PDF and try to select or search for the text you removed. If it's selectable, the redaction failed — this is the standard verification method for catching visual overlays that leave text extractable underneath the black box.

Then sanitize. Redaction removes what you can see; sanitization removes what you can't. Jeffrey Schoenberger's guidance for The Missouri Bar recommends running the "Sanitize and remove hidden information" step because it strips metadata, comments, bookmarks, attachments, form fields, and — increasingly important — embedded AI-generated metadata that may contain sensitive prompts or summaries even though it's invisible on the page. His framing is blunt: proper redaction protects clients, complies with confidentiality rules, and guards against sanctions or malpractice claims.

Keep three habits throughout:

• Work on copies, never originals — keep the source discovery file immutable so you can always show what you started from.
• Verify data handling before uploading client discovery to any free web tool; some process documents entirely in-browser while others upload them to cloud servers.
• Remember that automated detection is best-effort in every tool — human review of the output still matters, no matter which software produced it.

That last point deserves emphasis. Fully automated tools without a human review step can introduce compliance risk, and vendors themselves — not just bar associations — say so. AI-suggested regions are a starting point for review, never a substitute for it.

This is where the workflow gets hard with free tools. Search-and-redact features only work on PDFs with an OCR text layer, and the fully free open-source options trade security for searchability — pdf-redact-tools strips hidden text layers but produces output that isn't text-searchable at all. If your discovery is scanned police reports with no text layer, you're stuck doing manual rectangles.

Evidence Copilot handles this by OCR'ing scanned PDFs into searchable, citable text while keeping the bounding boxes needed for redaction. Redactions produce a new derived file — originals are never touched, every action is audited, and AI-suggested regions (faces, PII screens, license plates) file as cards marked "AI-suggested — needs review." Nothing is auto-applied; you confirm each one.

If you want to see the workflow end-to-end, book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan ($39/user/mo) at evidencecopilot.com/signup.