Security
Last updated
Your document stays in your browser
When you choose a PDF, the tool page hands it to a background worker (a Web Worker), which holds the document while you work on it. The PDF engine, MuPDF built for the browser, runs inside that worker too.
The tool page is a fixed file, so no server of ours takes part in reading your document. What the page does ask our own server, and what that carries, is in the privacy policy’s Your documents.
When it ends
Leaving the page, closing the tab or pressing Start over stops the worker, and its memory goes back to your browser at once. Opening a second document ends the first one’s session. Once that session has ended, nothing in the page can reach the first document, though its memory is only released for reuse until you leave the page or start over.
The cleaned copy waits in the page only until you download it.
Until the session ends, the page also keeps a handle to the file you chose, so it can read that one file again without asking. It stores nothing new: the file is already on your device, where you chose it.
What no website can control
Some things on your device are beyond any website, RedactNest included: your operating system moving memory to disk (swap or page files), your browser keeping memory or crash reports of its own, a browser extension that can read the page, and what is kept on your own disk.
Removed, not just covered
Drawing a black box over words hides them from your eyes, not from the file. The words are still there underneath, so copying the text brings them back.
A box drawn over the text
A copy of the text givesContact: jane.doe@example.com
Removed, then marked
A copy of the text givesContact:
RedactNest works the other way round: removed, then marked. It takes each item you ticked out of the text the page draws (its drawing instructions, the part of the file that says what text goes where), then draws the black box where it was. The box is a filled shape, not text, so there is nothing in it to copy.
Under each removed item it also blanks the pictures MuPDF reports there, over an area reaching a little past the item’s edges, and small drawn marks lying wholly inside that area, such as an underline, can go with it. Ink of a scanned page lying further out than that can stay visible. The check fails any file holding a figure or picture that no page’s drawing instructions reach, directly or through what they use. So the original copy of a figure or picture it redacts can stay only where a page’s instructions still reach it, such as on another page where you ticked nothing.
MuPDF does the removing, with every setting spelled out rather than left to its defaults, so a change in the engine’s defaults cannot quietly change what is removed. RedactNest never edits a page’s drawing instructions itself, and a ticked item MuPDF cannot remove stops the run.
The check on every file
Before you get a file, RedactNest opens the new file again, from the exact bytes you will download, and checks three things:
- Selectable text, spaces aside: for each page’s visible area, every character wholly inside it is still there, in its place, except those lying in an item you ticked, and none lying in one is left. Outside the visible area, none is left at all, and a character crossing the page’s edge can go with what lies outside it. It reads the pages twice, the second time ignoring replacement text (the words a copy gives instead of what is drawn).
- Pictures: the pictures the check sees under each removed item are blank there, and a picture it cannot decode counts as not blank. A picture drawn only as a repeating pattern, or used only as a soft mask (to fade or shape something else), can escape both the blanking and this check.
- Structure: the file holds a single version. The check follows the file from its starting point, and fails it if any object is left unreached, if any attached file is in it, or if an object holds an entry that is not on the fixed list for its kind (a font, a picture, a colour and so on), such as entries a program adds to the file’s objects for its own use. Almost every entry on those lists is one a page needs to draw. Names a file chooses, such as those it gives its fonts, pictures and colours, are held to no list: they can stay as written, and can carry words.
If any part fails, you get no file, only a plain reason. RedactNest stops rather than hand you a file the check could not vouch for.
Hidden data
The home page lists the kinds of hidden data.
Details stored inside a picture’s own data, such as a photo’s camera details, can stay.
That list leaves some things out. RedactNest also strips the file’s identifier, page labels, viewer preferences and automatic actions that are not scripts, and does not report these four. Links and accessibility tags are stripped too, so links in the cleaned copy no longer work and screen readers read it less well.
Notes and attachments
The home page notes how comments and form fields are flattened, and what goes with them. A sticky note or a file attachment is taken off the page, though one that shows as an icon can keep that icon as a mark on the page.
Passwords and fonts
- A file that opens without a password but restricts what you can do with it is not refused for that, and the cleaned copy carries no restriction. That is on purpose: you already hold the content, and the restriction protects nothing from you.
- A font in the file can still hold the shapes of characters that were removed. That can tell someone which characters the document used.
What it looks for
RedactNest looks for the kinds of detail the home page lists. It does not look for names or postal addresses, and you can add those as custom terms.
Each find is a row with a box. A row’s starting tick is a suggestion. While no run is under way, you can tick or untick any row except one RedactNest can tell, while reading, that it cannot remove safely.
Custom terms
A custom term is plain words, never a pattern or a wildcard. It matches its words through many differences of writing, such as case, accents, spacing and line breaks, but not all. What happens to the terms you type is in Your documents.
What it can miss
RedactNest goes by what the text looks like; it does not understand the document.
- A value broken up across lines, columns, table cells or pages can be missed, or found only in part. A term with other text between its parts, on one page or across pages, can be missed.
- One joined to other letters, digits or words can be missed, found only in part, or listed unticked.
- One written in a format, language or script RedactNest does not read, such as a phone number in another country’s own format, or a date written in words or with its month in another language, can be missed.
- Something that is not sensitive can be listed, and two card numbers side by side can share one row, so they stay or go together.
So look over the whole document before you share it, not only what was found and ticked (see the terms’ Checking the result is your job).
Pages it can’t fully read
RedactNest does not read words that exist only as a picture, so it cannot find them.
RedactNest’s page warnings rely on rules of thumb, so they can miss a page, and can warn about one that is fine, such as a page with a brochure photo or a line of very small print.
Some things can stay in the file, or be misjudged:
- When a scan carries text from recognition (OCR) run by another tool, RedactNest can treat the picture as read and work from that text alone, so words the recognition missed or misread can stay in the picture.
- It can name a scan as a picture it cannot read although recognised text lies over it.
- The part of a picture beyond the page’s visible edge can stay in the file.
- Text a clip hides completely can stay in the file without a row.
- Drawings outside the visible area can stay in the file.
- Descriptions written into a page, such as alternate text for screen readers, can stay in the file, even around text you tick. RedactNest looks for them and warns about each page where it finds some.
Fake redactions in files you receive
RedactNest looks for text under a box or picture drawn over it, and for text you cannot see, such as text the same colour as the page. When it finds some, it warns you that the text is still in the file. This search can miss hidden text, so a file you received can still hold words that look covered, with no warning.
When it stops instead
At open, RedactNest refuses, among others, a file that is not a PDF (judged by its contents, never its name), one too damaged to open, one that needs a password (RedactNest does not take passwords), one with layers a viewer can switch on and off, and a file in which it finds no page it can read.
Some items cannot be removed safely, such as text set at an angle or inside replacement text. Where RedactNest can tell while reading, the row says the item will stay in the file.
Otherwise the run stops with a plain reason and no file, and your ticks are kept: a row you could tick can still stop the run, for something RedactNest meets only while removing. Some files stop every run. A run can also end because the engine itself was lost. The review then starts over, without your ticks, when you press Try again.
Plans and the page limit
How the tool page learns which plan applies to you, and what that question carries, is in Your documents.
The page limit relies on honesty. It is applied in your own browser, by code anyone can read, so someone determined could lift it. Pro gives you longer documents and custom terms, not a lock, and custom terms rely on honesty in the same way.
Logs, counts and error reports
On purpose, there is no record of your document or its words, because such a record would recreate the very exposure RedactNest exists to remove.
MuPDF’s own messages, which can quote a document, are thrown away inside the worker as the engine loads. Whether our own code writes a log is in What we do not do, and what our host keeps is in What we receive when you visit.
What we count, what a failure report holds and what neither ever holds is in Counts and error reports, and which outside scripts run on our pages is in What we do not do.
Check it yourself
The foot of every page links to the source code for the version you are using.
Your browser can show you the rest. Open its developer tools, then load the tool page and redact a file, and the network panel lists every request the page sends, so you can check what Your documents says.
If you find RedactNest doing something this page says it does not, please write to privacy@redactnest.com.
Changes
When we change this page, we update the date at the top and add a line below saying what changed.
- : The original copy of a redacted figure or picture can no longer stay in a font character. RedactNest now stops, at open and again before it removes anything, on a file whose font characters draw a figure or picture the rest of the file also uses, or draw text of their own.
- : The check now fails a file holding an object it does not reach from the file’s starting point, or an entry off the fixed list for that object’s kind, though names a file chooses can stay as written and can carry words. So this page says where the original copy of a redacted figure or picture can still be, what else RedactNest strips, and that descriptions written into a page can stay, with a warning on each page where RedactNest finds some. The check’s text part now says it reads selectable text. What it finds is now What it looks for.
- : First published.