← Secure tools

Guide

PDF metadata and hidden information

The pages you see in a PDF viewer are only part of the file. Metadata, earlier revisions, attachments, form fields, scripts, and links can travel with it. This page explains those layers and the limits of a local scan.

Layers that are easy to miss

The Info dictionary and XMP can store title, author, creator, producer, and dates. Incremental updates can leave an older body in the file even when the viewer shows the latest pages. Embedded files are attachments. AcroForm fields can hold names and answers. JavaScript actions, open actions, and launch actions can be present even if you never saw a script run. URI actions can point at sites you did not intend to advertise.

A redacted-looking black box is sometimes only a drawing on top of text. This scanner does not decide whether text is visually covered. It also does not OCR scanned pages.

What a local scan can report

The PDF tool accepts a PDF up to 32 MB and parses it in a same-origin worker. It can report author and producer metadata, JavaScript and other actions, external URIs, forms, and embedded files. Truncated previews avoid copying a long secret query string into the result card. A malformed file can still yield a partial finding, such as a leftover author string, together with a truncated-file warning.

Encryption is reported as a limit on inspection. If the scanner cannot see inside an encrypted object, absence of other findings is not reassuring.

When cleaning is refused

Metadata cleaning is intentionally narrow. The cleaner will not rebuild a PDF when active content, attachments, forms, encryption, revision history, external links, compressed-object limits, or malformed structure are present. That refusal is the safe outcome. Rebuilding those files incorrectly can drop pages or hide the fact that a script was there.

For a finding the cleaner will not touch, export a new PDF from the source application without the extra feature, or delete the attachment and scripts in an editor, then scan the new file. Do not flatten by taking screenshots if the text still needs to be selectable, unless you have decided that tradeoff on purpose.

How to read a result

Treat each finding type as a question. PDF_AUTHOR asks whether that name should travel with the file. An embedded-file finding asks you to open the attachment separately. A JavaScript finding asks you not to rely on a metadata wipe. Zero findings mean this parser did not see the patterns it implements.

A fictional example

A project PDF shows Author “Jane Doe”, a title, creator, producer, and an XMP date. Another file in the same folder has an open action, a JavaScript action, and a URI whose query contains an internal id. A third file has an AcroForm and an embedded spreadsheet. The scanner should report those as separate finding types. Only the first file is even a candidate for metadata cleaning, and only if none of the refusal conditions are present.

The names and the internal URL are illustrations. Do not paste a real customer PDF into a shared bug report. Scan it locally, and if you need help, describe the finding types rather than the document.

Frequently asked questions

Quick answers from this guide.

Does the scanner remove hidden text behind a black box?

No. It does not judge whether text is visible. Assume covered text is still in the file until a PDF tool you trust rewrites the content stream.

Why is the link preview cut off?

So the results list does not become a second copy of a secret query string. Open the original only on a machine where that URL is supposed to exist.

Is a 32 MB limit a product outage?

No. Larger files are rejected so the tab does not try to parse an unbounded document. Split or shrink the PDF, then scan the part you actually intend to share.

Use the related tools