Sandbox reading
The document's JavaScript, open actions (OpenAction), external commands (Launch) and links are not executed, and the window has no network access: active code is extracted and shown, not run.
PDF documents
PDF workbench
A PDF that lands in your inbox, that you download or that a client hands you is not a sheet of paper: it is a format with a programming language inside. Opening it with Acrobat or Preview means running it — its JavaScript fires, open actions are followed, and the document can reach the outside and tell whoever crafted it that it was opened, when and from which IP address. Probatio's PDF module instead opens it in its own window and in a multi-layer sandbox: the document is interpreted, not executed; the window has no network; the raw bytes are parsed in a subprocess isolated from the rest of the system, with a time limit and — on macOS — also memory and CPU limits and an operating-system profile denying it network and any write outside its work folder. It is a safe reader for anyone who has to open a suspicious attachment — analyst, professional, cautious user — and at the same time a forensic workbench to examine it (provenance, active code, hidden text, signature) and work on it (pages, redaction, annotations, stamp) without ever touching the original.
Security · Defence in depth
Three layers between the PDF
Three layers between the PDF
and your machine
A PDF is not a sheet of paper: it is a format with a programming language inside. Opening a document you don't trust is, quite literally, running code from a potentially hostile source. Probatio does not just «turn JavaScript off»: it separates what the document asks to do from what the system lets it do, at three independent levels — if the first falls, the other two remain.
-
The document is not executed — Rendering engine with
evaldisabled, scripting off, no active annotation layer: JavaScript,/OpenAction,/Launchand links do not fire. Active code is extracted and shown. - The window cannot get out — A restrictive Content Security Policy limits connections to the app itself: no «phone home», nobody learns that you opened the file.
-
Parsing lives outside the application — The file's bytes are read by a separate helper (
probatio-pdf-helper) with a timeout and, on macOS, with a memory and CPU cap. A malformed PDF brings down the helper, not Probatio, and does not take the working session with it. -
On macOS, the OS itself enforces confinement — The helper runs under a
sandbox-execprofile that denies network, denies writes outside the work folder and denies spawning other processes. Tested: an attempted write to/etcis refused by the kernel, not by an in-app check. - The file's bytes never leave the sandbox — The helper only produces derived results (metadata, decoded images, cleaned copies). Delivery to a path you choose is performed by the trusted main process: the sandbox never writes where you decide.
- The original is never touched — Each window works on a session copy. Undo and redo are full document snapshots, not inverse operations: an «undo» that sometimes does not undo is worse than no undo at all.
How far it goes, and where it stops. Full OS-enforced confinement is active on macOS (a sandbox-exec profile): no network, no writes outside the work folder, no child process, memory and CPU limits. On Windows it is partial: the helper runs in an operating-system Job object, with limited memory and CPU and unable to start other processes, but network and disk writes are not denied yet. And no sandbox makes a file harmless: basic sanitizing removes JavaScript, automatic actions and attachments, but an exploit nested in a malformed stream or in a font would survive. To go after that very vector Probatio offers a deep clean: it rebuilds the document, discards embedded font programs (remapping fonts to a standard one) and neutralizes dangerous-codec images — JBIG2 and JPX, the class of the FORCEDENTRY attack. Both sanitizing and deep cleaning are always your choice, declared, at save time — and when the PDF is evidence to be filed, keep in mind that cleaning it alters it: the cleaned copy is no longer the original.
What it does
What the PDF workbench module does
Parsing isolated from the system
The PDF bytes are read in a separate subprocess, with a timeout, a memory cap (1.5 GiB) and a CPU cap (30 s for each processor it may use, up to 16): a malformed or malicious file can bring down the helper, not the app. On macOS an operating-system profile denies it network, writes outside its work folder and spawning other processes; on Windows it runs in a Job object that limits memory and CPU and stops it from starting other processes.
Provenance inspection
The program that generated the PDF, creation and modification dates, source file (xmpMM:DerivedFrom), inferred operating system from the producer string — labelled as an inference, because no PDF records its own OS — and the full XMP packet.
Extracted active code
Every object in the document is walked looking for action dictionaries: JavaScript, Launch, URI, SubmitForm. The code is read in full in the Inspection column. Looking only in the «well-known places» would always miss one.
Hidden text and poisoning
White-on-white text, in invisible render mode or in a microscopic size: invisible on the page, but read by whoever extracts the text — automated review, indexing and AI models. Probatio lists it page by page and takes you to it with a click.
View & search
Zoom, in-text search with match highlighting, page thumbnails, single- or two-page view and a split view (vertical or horizontal) to compare two distant pages of the same document.
Pages, undo and redo
Delete, rotate, reorder and insert pages from another PDF, by dragging it to the chosen spot among the thumbnails: imported pages arrive complete, with the attributes inherited from the source document. You always work on a session copy, with undo and redo on full document snapshots.
Real redaction
Redaction removes the glyphs from the file, it does not put a black rectangle over them: the redacted text is no longer extractable by copy-and-paste nor by pdftotext. Images and annotations inside the area go too; the document is re-serialised, so the previous revision does not survive inside the file.
Annotations
Highlights in five colours, notes (sticky) and arrows. They are real PDF annotations, each with its own appearance stream: Acrobat reads them, Preview reads them, and they can be removed without touching the content.
Stamp & image library
Apply an image (e.g. a signature PNG with transparency) and resize it with locked proportions. Embedded images sit in a library with thumbnails and a lightbox, zoomable 1:1 at their true resolution: JPEGs are saved bit-identical to the embedded stream, so the extracted image's hash matches.
PAdES signature verified on the original
A badge flags whether the PDF is signed; on the first edit it warns that the signature will be invalidated. The detail (validity, integrity, signer, eIDAS/QSCD, chain, timestamps, OCSP) is checked on the original file, not on the work copy: verifying the copy would tell a lie about the document you received.
YARA and informed saving
On opening, a YARA scan warns about code known to be suspicious — and if the rules cannot be downloaded it says so: «I could not look» is not «clean». When saving a document with active code, you choose between a sanitized copy and the original file as it is.
Printing without opening it elsewhere
Pages are drawn by the same sandboxed engine as the view and handed to the system print dialog: the PDF is never passed to another program, so printing a suspicious document does not mean opening it in Acrobat or Preview. You print what you see, edits and redactions included.
Selectable text
PDF text can be selected and copied as in a normal reader, without the document running anything.
Saved versions and comparison
A PDF saved several times keeps its earlier versions: Probatio lists them with date, program and the signatures covering each, opens them as separate files and compares them with the current one, page by page, with a map of the differences. It is the way to see what changed after a signature.
Layers, attachments, fields, outline
Layers on and off (including those hidden on opening), attached files with their hash, form fields exportable to CSV and JSON, outline and links (never opened), embedded fonts and images with the EXIF metadata left in JPEGs.
Continuous scrolling
Pages scroll one below the other, and the most visible page is the one the tools work on; “Page by page” stays in the Layout menu. Thumbnails and the Inspection column can be closed and the choice is remembered; the toolbar sits on two fixed rows, the document above and the tools below.
PDFs written by older tools
PDFs from Crystal Reports and similar tools, with internal tables written off-standard, now open: Probatio corrects the reading in memory, without touching the file.
Step by step
How it works
- Drag a PDF onto the module — or use “Open with Probatio” from the OS context menu. It opens in its own window.
- The sandbox interprets it without executing its JavaScript and without network, while byte parsing happens in an isolated subprocess. Active code is shown, not executed.
- Inspect provenance, active code, hidden text, embedded images and the signature; check the YARA matches in the Inspection column.
- Work on the pages, redact, stamp and annotate: always on a copy, with undo and redo.
- Save as: the original is never touched. For a suspicious PDF choose the sanitized copy, knowing that sanitizing does not guarantee harmlessness and that the cleaned copy is no longer the original.
FAQ
Frequently asked questions
Why is opening a suspicious PDF with Acrobat or Preview risky?
Because a PDF is not a sheet of paper: it contains a programming language. A normal reader executes the document's JavaScript, follows its open actions (OpenAction) and can reach the outside — telling whoever crafted the file that it was opened, when and from which IP address.
What does it mean that Probatio opens the PDF «in a sandbox»?
It means three independent layers. First: the rendering engine has eval and scripting disabled, so JavaScript, OpenAction, Launch and links do not fire. Second: a restrictive Content Security Policy prevents the window from making any network connection. Third: parsing of the raw bytes happens in a separate subprocess, with a timeout, and on macOS with a memory cap (1.5 GiB), a CPU cap (30 s per processor, up to 16) and under an operating-system profile denying it network, writes outside its work folder and spawning other processes.
Does the sandbox work on Windows too?
Partly. On Windows the first two layers apply — no script executed, no network from the window — and parsing runs in a separate subprocess with a time limit, so a malformed PDF brings down the helper and not the application. OS-enforced confinement (no network and no writes outside the work folder for the helper, memory and CPU limits) is currently active only on macOS: not yet on Windows.
Does the sandbox make a malicious PDF harmless?
No, and no honest tool promises that. The sandbox limits what the document can do to your machine; basic sanitizing removes JavaScript, automatic actions and attachments. For the most insidious vector — an exploit nested in a malformed stream or in a font — there is a deep clean that rebuilds the document, discards embedded font programs and neutralizes dangerous-codec images (JBIG2/JPX). Even so there are no absolute guarantees: this is why the choice to clean always stays with you, declared, at save time; and when the PDF is evidence to be filed, keep in mind that cleaning it alters it and the copy is no longer the original.
What is hidden text in a PDF?
It is text present in the file but invisible on the page: white on white, in invisible render mode or in a microscopic size. You do not see it while reading, but anyone extracting the text does — automated review, indexing and AI models — and it is the classic route for prompt injection or poisoning. Probatio lists it page by page and takes you to it with a click.
Does redaction really remove the text from the PDF?
Yes. Probatio removes the glyphs from the content stream and replaces them with an equivalent advance, so the remaining text does not shift; it also removes images and annotations contained in the area, and only at the end draws the black rectangle. The document is re-serialised, not incrementally updated, because an incremental update would keep the previous revision with the original text. The limits (glyphs with uncertain metrics, images intersecting the area without being contained in it) are declared and counted, not hidden.
Does editing a signed PDF invalidate its signature?
Yes, and Probatio tells you beforehand: a badge flags that the document is signed, and the first operation that modifies it raises an explicit warning. Signature verification (validity, integrity, signer, eIDAS/QSCD, chain, timestamps, OCSP) is always performed on the original file and not on the work copy, which may already have been modified.
Is the original modified while I work?
Never. Each window duplicates the document into a session folder and every operation acts on that copy; the original stays untouched on disk until you choose where to save the result. Undo and redo work on full document snapshots, so an undo always truly undoes.
Related modules