Privacy risk scanner
Scan a PDF for unapplied /Redact annotations (marked but not removed: a real black rectangle drawn in a page's own content stream with no annotation at all is not detected) and hidden optional-content layers, or a .docx for tracked changes, hidden text runs, and the author names and user ids left in word/*.xml (word/people.xml included, where Word keeps a comment author's name and e-mail address). Reports what was found, not a verdict. "Clean" here means these specific checks found nothing, not that the file is safe to share. Cropped-but-embedded image data is explicitly not checked (see the tool source for why).
Runs on your device. The file is never uploaded.
Privacy risk scanner lists a PDF's optional content layers, flags those hidden by default, and names any page whose /Redact annotation still has live text under it. A .docx instead gives tracked-change counts, hidden runs and every author name left in word/*.xml. A rectangle painted into a content stream carries no annotation, so that redaction goes undetected.
Questions
What does it check in a PDF?
Two things. It lists optional content groups, the layers, and flags which ones are hidden by default, since content under a hidden layer is still fully embedded. And it looks for a /Redact annotation on a page that still has extractable text, which is the signature of a redaction that was marked but never applied.
Will it catch a black box drawn over text?
No, and it says so in its own output. A rectangle drawn directly into a page content stream, the way markup tools do it, carries no annotation for the scanner to find, and that is the most common fake redaction of all. This tool detects the unapplied /Redact case, not that one.
What does it check in a docx?
Counts of tracked insertions and deletions, the number of hidden text runs, whether a comments part exists, and every author name and user id still present in the Word XML parts. That last one includes the people part, where Word 2013 and later store a commenter name and, in a managed tenant, their email address.
Does a clean result mean the file is safe to share?
No. It means these specific checks found nothing. The tool reports findings rather than a verdict, and it names one check it explicitly does not perform: cropped but embedded image data, which needs clip path and transform geometry analysis this build does not attempt.
Which formats does it accept?
PDF and .docx only, chosen by file extension. Anything else stops with a message saying so. It pairs with the Universal metadata scrubber, whose own output tells you to run this on the scrubbed file to see what is still there.
Is my document uploaded to be scanned?
No. Both paths run in this tab: pdf-lib and pdf.js read the PDF, and the docx is unzipped and searched in memory. Nothing is transmitted, which is the reason to scan a sensitive document here rather than on a server you do not control.