Signatures are personal information in some releases and routine in others, and the rule that decides which usually comes from records law or a court rule rather than from anything a redaction tool can know. The question tends to arrive with a signed contract, an affidavit or a notarized deed that is due out with a release, and the practical work turns out harder than the question suggests. Forms, signature blocks and stamps defeat tools built for running prose, because a signature is a picture and a form keeps its values apart from its page. Our guide to removing content from a PDF for real covers what removal must achieve in any document. Forms add a policy question to that technical work, and the policy comes first, since it decides which signatures need finding at all.
Should signatures be redacted?
Court rules and records laws answer the question differently, and the difference explains why it keeps coming back. In federal court, the privacy rule for filings limits what a filing may include from a handful of sensitive identifiers, and signatures are not among them. Public-records law can point the other way, at least for the signatures of individual employees. In its chapter on Exemption 6, the Department of Justice's guide to the Freedom of Information Act cites an unpublished 2017 Ninth Circuit decision on exactly this point, American Small Business League v. Department of Defense. The court held "that DOD properly withheld business contact information and signatures of employees whose privacy interests were small, but not trivial, as this information could be used for harassment or forgery."
A sound approach is to treat every signature as personal information, decide by written policy which ones are withheld, and record the basis on each redaction. A policy might withhold the signatures of private individuals and junior staff while releasing those of officials who signed in an official capacity. Printing the exemption code on each redacted signature lets a requester see why it is missing. None of this is legal advice, and the policy belongs to the organization's counsel.
Why text search never finds a signature
A handwritten signature is a picture, whether it was signed in ink on paper that was later scanned or pasted into a digital document as an image. Keyword search and text-based detection look for characters, so they pass over a signature without registering it. The one exception is a typed signature line, such as the "/s/" followed by a name used in many electronic filings, which is ordinary text and turns up like any other name.
Finding signatures therefore takes object detection that runs on the rendered page and looks for the shape of a signature, the same way detection finds a face in a photograph. Detection of that kind proposes regions, and a reviewer confirms them page by page, because a flourish in a letterhead or a scribbled note can resemble a signature and a faint signature can go unnoticed. Initials repeated on every page of a contract need the same attention as the signature block, since they identify the same people. Handwritten notes near a signature are a separate problem, covered in our article on handwritten and rotated scans.
Fillable forms: flatten a copy before redacting
Values typed into a fillable PDF, and the text of its comments, live outside the page's drawing instructions, so a redaction that filters only what the page draws leaves them in the file. Our article on the hidden places PII survives tells the case of a real form whose values sat in three places the page never showed.
Redacting a fillable PDF safely comes down to a short sequence that works with ordinary tools. Keep the original, and flatten a copy so that every field value and comment becomes part of the page. Then redact the flattened copy and confirm that no field in the released file can still be typed into. The released copy is then no longer a fillable form, which suits a released record, since a requester needs to know what the form said and has no use for a form they could edit.
How form layouts trip up automated detection
Scanned forms break the assumptions that detection built for prose relies on, and we have worked through several of them on real forms. Detection of a name or a date of birth often leans on a label such as "Name" or "DOB" printed before the value. On a scanned form the label can sit after the value or below it, and OCR can read the page out of order, so a detector that only looks backward from a value misses it. VIDIZMO Redactor looks for label words on both sides of a value in scanned text.
Account and identity numbers fail in a different way when a form squeezes them into narrow boxes. A number that wraps onto a second line defeats a pattern that expects one unbroken number, which then matches nothing, or matches and masks only the half it recognized. Redactor treats a match that spans a line break as two pieces and masks both halves.
Two values printed on one line cause a different problem, one that depends on how the reader groups the text it finds. A provider identifier and a phone number printed on the same line become a single region if the reader returns whole lines, and the redaction must then cover the entire line or leave it alone. Boxing each word separately lets each value be masked on its own. Where a scan runs two fields together with no space between them, the fused word is masked whole instead of being cut at a guessed boundary.
A ticked checkbox beside a diagnosis or a benefits category reveals sensitive information without containing any text at all, so neither text search nor text-based detection will ever flag it. A checkbox in a fillable form is stored as a field value, so flattening brings its state onto the page with everything else. On a scanned form, a tick is only a mark in the image. Reviewers of health, benefits and employment forms should look at every checkbox on a page and draw regions over the ones whose meaning must be withheld.
Stamps, seals and notary blocks
Many stamps carry exactly the details a release has to withhold, even though software that searches for text sees them only as graphics. Under California law, for example, a notary's seal must show "the name of the notary," the county where the notary's bond and oath are filed, and "the date the notary public's commission expires." Received stamps and date stamps carry names, dates and sometimes case numbers too.
Embossed seals have little contrast on a scan, rubber stamps are often faint or overlap signatures and text, and their shapes vary from one office to the next. The dependable approach is to find them during review and draw the regions by hand, paying particular attention to notarized contracts, deeds and affidavits. Our article on redacting loan and mortgage documents covers notarized closing packages in more detail.
Can you redact a signed PDF?
A digitally signed PDF can be redacted, but the digital signature will not survive the redaction, and a signature that did survive would be a warning sign. Researchers who studied signed PDFs explain that "A typical signature starts at the first byte and ends at the last byte of the trailer," so the signature covers essentially the whole file as it was signed. Later changes are appended as updates, and the same researchers note that "Since such Incremental Updates are appended to the signed document, and no changes within the signed area are made, the signature remains valid."
A redaction that kept the signature valid would have to leave the signed bytes untouched, which means leaving the unredacted original inside the file. A true redaction rewrites the file, and the signature breaks as a result. The sound practice is to redact a copy, keep the signed original under access control as the record, and log who produced the redacted copy, from which original and when. Anyone who needs to verify the signature can then be given access to the original instead of the release. Opening both files in a viewer that validates signatures confirms which is which, since the original should still validate and the redacted copy should show no valid signature at all.
Documents signed through an e-signature service often carry one more page to check, a completion certificate or audit trail that lists each signer's email address, IP address and signing times. That page needs the same review as the signature block itself, and it is easy to miss because it sits at the end of the file after the pages everyone reads.
Signatures and forms in a Redactor job
Redactor detects handwritten signatures as objects on the rendered page, for redaction or for review, and flattens form values and comments into the page before removing anything. Regions an analyst draws over stamps, seals and checkboxes are removed exactly as detected ones are.
Records teams deciding which signatures to withhold can see how exemption codes and review fit a release on the FOIA redaction software page.