Privacy Mask: Hide Sensitive & Personal Info (PII) by PaperWork

Hide sensitive and personal information in text before sending it to AI assistants. Processing stays local in your browser.

Privacy Mask hides personal and sensitive information in your text before it reaches an AI chat. It runs on your own computer. No tracking, and detection never sends your text anywhere. Start on the Free plan with no account, or go Pro for unlimited hides and files. You stay in charge the whole time. Privacy Mask never sends a prompt for you and never changes your text silently. It shows you what it found, lets you choose what to hide, and waits for you to insert the masked text and press enter yourself. HOW IT WORKS Write or paste your prompt the way you normally would. Open the Privacy Mask panel and it scans the text right away, on your device. You get three things side by side: the original, a masked preview, and a list of every item it detected. Turn individual items on or off if it caught something you want to keep, or missed something you want to add. When the preview looks right, insert the masked version into the chat box. Each piece of sensitive data becomes a short placeholder, such as [EMAIL_1], [NAME_2], or [PHONE_1]. The link between a placeholder and its real value lives only in your current browser session. So when the AI replies, you can paste the answer back and Privacy Mask restores the real values locally, which keeps the reply readable for you. Close the session and that link is gone. PLANS Privacy Mask comes in two plans. Detection works the same way in both, locally in your browser; the plans differ only in how much you can hide each week. Free: 200 hides and 10 files per week, no sign-in needed. The counter lives in your browser and starts over seven days after the first hide. Pro: unlimited hides and files. $9 per month or $60 per year, cancel any time from Manage subscription. Pro needs a Google sign-in so the plan follows you between computers. Stripe takes the payment, so no card details reach PaperWork. Signing in sends only your account email and plan status to PaperWork's account service, never anything you hide. WHERE IT WORKS Privacy Mask adds a quick privacy step to the major AI chat assistants you already use. The full list of sites it runs on is shown in the permissions on this page, so you can check it before installing. It activates only on those sites and stays completely inactive everywhere else. On your bank, your email, or any other page, the extension does nothing at all. The document and text editors are separate. They open in their own extension tab, work with no AI chat involved, and need no network connection at all. WHAT IT DETECTS Privacy Mask looks for the kinds of data that cause real trouble when they end up in a prompt. Identity and contact details - Names of people and of business entities - Email addresses - Phone numbers, validated across many countries rather than a single region - Postal and street addresses Financial and banking data - Credit card numbers, checked with the Luhn algorithm - IBAN and BIC / SWIFT bank codes - Bank account numbers Government and tax identifiers - National and tax IDs for the United States, United Kingdom, Spain, Italy, France, Germany, the Netherlands, Russia, and the United Arab Emirates, each confirmed by its own checksum - Passport numbers - VAT numbers and the UAE Emirates ID Secrets and credentials - API keys and access tokens from common developer and cloud platforms - Database connection strings that carry a username and password - Bearer tokens and private keys Technical and other identifiers - Crypto wallet addresses - IP and MAC addresses - Vehicle VINs and license plates - Dates and money amounts BUILT TO AVOID FALSE ALARMS A tool that masks everything that vaguely resembles an ID becomes useless fast. Privacy Mask works in two steps. A pattern finds a candidate, then a check confirms it. Credit cards pass the Luhn test. IBANs pass the mod-97 calculation. National IDs pass the checksum rules for their country. A random 11-digit order number sitting in your text is left alone, because it fails every one of those checks. Numbers that only mean something in context, like a tax ID, are masked when the words around them make the meaning clear, and ignored when they do not. DOCUMENTS: PDFS, SCANS AND IMAGES Sensitive data does not live only in typed prompts, so Privacy Mask includes a document editor. Open it from the side panel and load a PDF, JPG or PNG. You see the real pages and a list of what the rules found: type, value and page. - Every occurrence has its own Hidden / Visible switch. Two mentions of the same name are independent, and "Apply to all" changes them together only when you ask for it. - Draw a box over anything the rules missed, or select text and choose Hide selected text. Undo and redo cover both. Original and Result show the page before and after. - Export does not draw a black rectangle over the text. It renders each page, erases those pixels, and rebuilds the PDF from the cleaned images. No original text object, annotation, form field, attachment or metadata survives. The result is not searchable, and the editor tells you so. - The default file name has the hidden values stripped out of it, and the exported file carries no table of originals. - Download PDF writes the file to your computer. Nothing is sent anywhere and no chat is touched. - Attach to... is a separate, explicit action: you pick an open chat tab yourself and hand over the current verified version. It is never sent for you. READS SCANNED PAGES TOO A scan is a picture, and a picture of a contract hides the same data as a typed one. Check page images runs text recognition locally on every page, including scans and mixed text-and-image PDFs. English, Russian and Spanish recognition data ship inside the extension. No download, no server, no optional model required. Recognition is not perfect and the editor does not pretend otherwise. Unchecked, empty and low-confidence pages stay marked as review-required. Review every page and use manual boxes for anything the recognizer missed. OFFICE FILES Word documents (.docx), spreadsheets (.xlsx) and CSV files go through the same review flow. The text inside them is scanned with the same detectors, and images embedded in those files can be checked as well. YOUR OWN TERMS Some sensitive words are specific to you: a project codename, a client alias, an internal product name. You can add your own terms, and Privacy Mask masks them next to the built-in detectors. Your custom terms are stored locally, like everything else. OPTIONAL: A LOCAL AI MODEL Rules and checksums handle the structured data well. For free-form writing, where a name or an address is defined by context rather than by shape, there is an optional local model called Privacy Filter. It is genuinely optional. Masking in chat, PDFs and Office files uses the built-in rules and works fully without it. If you never install the model, Privacy Mask downloads nothing after install. If you do install it, this is what happens and it is worth reading before you click. The weights come from Hugging Face and are close to a gigabyte, downloaded once. You can also point the extension at a copy you already have on disk instead. Once installed, the model runs inside your browser. Your text is never uploaded to it, because there is nothing to upload to: the model is on your machine. You can compare what the rules found against what the model found, side by side, and decide. WHY LOCAL-ONLY MATTERS Many tools that promise privacy do their scanning on a server. To check your text, they first have to receive your text, which is the exact thing you were trying to avoid. Privacy Mask takes a different route. All detection happens in your browser, using pattern matching, checksums and, if you chose to install it, a model stored on your own disk. Your text never reaches a backend that could be breached, nothing you type is logged, and nothing is sold. If you want proof, open your browser's network tab while you work. Detection produces no outbound requests. There are exactly two moments when anything leaves your machine, both started by you and neither involving your prompt text: downloading the optional model, and attaching a cleaned document to a chat tab you picked yourself. LANGUAGES Three different things use languages here, and they are set separately. - Interface: English, Russian and Spanish. - Detection by format: language-independent. Cards, IBANs, API keys, crypto addresses and national IDs are found by their shape and checksum no matter what language surrounds them. This includes the Spanish DNI, NIE and NIF, the Italian codice fiscale, the German and Dutch tax numbers, and the rest of the supported countries. - Detection by context, where a value is only identifiable from the words next to it, such as a name after "Client:" or a tax number after its label: English and Russian today, with more planned. - Recognition of scanned pages: English, Russian and Spanish. You pick this per document, separately from the interface language. WHO IT HELPS People paste real, sensitive material into AI tools every day: - Support agents pasting tickets full of customer names, emails, and order details - Developers pasting code, logs, and config files that still hold API keys and connection strings - Finance and operations teams pasting invoices, bank statements, and contracts - Anyone drafting an email or document that names a real person Privacy Mask lets you keep using AI for this work while keeping the personal and financial details to yourself. ABOUT PAPERWORK PaperWork builds document AI for banks, lenders, and fintech companies. Our main product reads financial documents such as bank statements, invoices, and KYC files, and turns them into structured, verified data. Handling that information carefully is the center of what we do, not a footnote to it. Privacy Mask grew out of the same concern. We wanted a way to use everyday AI tools without exposing the personal and financial data we are trusted to protect, so we built one. PaperWork sits where AI, financial technology, and data privacy meet, and this extension carries that focus into daily use: keep the help that AI gives you, without giving away the data behind it. You can learn more about our work at paperwork.to. WHAT PRIVACY MASK DOES NOT DO - It does not send your prompts anywhere. You insert and send them yourself. - It does not collect, store, or transmit your text off your device. - It does not run analytics or telemetry. - It does not load or execute remote code. The optional model download fetches data files only, verified against known checksums. - It does not require the optional model. Everything needed for rule-based masking ships inside the extension. A NOTE ON PERMISSIONS Privacy Mask asks only for what the job requires. - Access to the supported AI chat sites, so it can read the input field, scan your prompt and put the masked version back. - Local storage for your settings and custom terms, and session storage for the placeholder map. - Unlimited storage, because the recognition data and the optional model are far larger than the small default quota. - A local offscreen step, used to process PDFs and images away from the page. - Access to huggingface.co, used for one thing only: downloading the optional model when you ask for it. If you never install the model, this access is never used. - Identity, used only to open the Google sign-in window when you choose the optional Pro plan. The Free plan never uses it. None of this sends your text to a server we run. The only PaperWork service is the optional Pro account, and it sees your email and plan status, never your text. GETTING STARTED Install the extension, open one of the supported AI sites, and write a prompt. Open the Privacy Mask panel, review what it found, and insert the masked text. That is the whole flow. On the Free plan there is nothing to sign up for and nothing to set up before it works. For documents, open the editor from the panel and load a file. COMMON QUESTIONS Does Privacy Mask send my text anywhere? No. Detection runs in your browser. No request leaves your machine to scan your text. Will it submit my prompt for me? No. It only inserts the masked text into the input field. You read it and decide when to send. What happens to the link between [EMAIL_1] and the real email? It stays in your current browser session and is used to restore values in the AI's reply. When the session ends, it is cleared. Do I have to download the model? No. Rule-based masking is the default and works on its own. The model is an extra you can install, skip, or remove later. Does the document editor really remove the data, or just cover it? It removes it. Pages are re-rendered with the hidden pixels erased and the PDF is rebuilt from those images, so the original text is not recoverable from the file. Does it work in languages other than English? Partly, and it is worth knowing where the line is. Anything with a fixed format, including cards, IBANs, API keys and the national IDs of all supported countries, is found regardless of the language you write in. Detection that depends on surrounding words currently understands English and Russian. The interface is English, Russian and Spanish, and scans can be recognized in those three. How much does it cost? The Free plan covers 200 hides and 10 files a week with no account. Pro is unlimited for $9 a month or $60 a year, with a Google sign-in and payment through Stripe. You can cancel any time from Manage subscription. Privacy Mask comes with a Free and a Pro plan, and the way it handles data is set out in full in our privacy policy. If a claim on this page matters to you, you can check it yourself, because the work happens where you can watch it: in your own browser.