hiddentokens

Hidden Tokens / Academy / Lesson 2

Protect a folder and check the coded copy

Lesson 2 of 9 · 5 min · app 0.14.4 · connector 2.42.0

How do I anonymise a Word or Excel file before an AI reads it?

Put it in a folder that Hidden Tokens watches: within seconds, a coded copy appears in the folder's Hidden-Tokens subfolder, with names, identifiers, bank details and dates of birth replaced by codes, headers and comments included, and that copy is the one you give the AI. Then check it yourself: a search for each real name, phone number and IBAN in the coded copy should find nothing, while an amount in the body of a contract stays readable by design.

Before you start

You need the Hidden Tokens app on a Mac with Apple Silicon (macOS 13 or later) or on Windows 10 or 11, 64-bit. There is no version for Intel Macs. The 14-day trial starts with your first coded file, with no account and no card.

On Windows, the installer is not signed with a Windows certificate yet: on first launch, SmartScreen says "Windows protected your PC". Click More info, then Run anyway. The app installs for your user only, without administrator rights.

Step by step

  1. Create an empty practice folder

    Install Hidden Tokens from the download section if you have not yet (Mac with Apple Silicon, or Windows 10 and 11). Then create an empty folder, for example Academy practice in your Documents. Unzip the Academy sample folder somewhere else, on your Desktop for instance: you will drop its files in a moment.

  2. Watch the folder

    Click the Hidden Tokens icon in the menu bar (Mac) or the notification area (Windows), choose Watch a folder…, select Academy practice and click Watch. The app confirms with Folder watched. On a Mac, macOS may ask whether Hidden Tokens can access your Documents folder: accept, it is the app that reads your files, not the AI.

  3. Set the code language to English, before the first file

    From the same menu, choose Open dashboard. On the card of your folder, open Settings and set Code language to English. Do it now: the language is fixed by the folder's key at the first coded file. With English, the codes read NAME-001, and English letters are read as English.

  4. Drop the four files into the folder

    Drag the four files of the sample folder into Academy practice. Within about ten seconds a Hidden-Tokens subfolder appears, with the coded copies: their names end in -anonymise. The letter gives a coded .docx (and a .md text version), the payroll a coded .xlsx, the PDF contract a coded .md. Your original files are never modified.

  5. Search the coded letter for what should be gone

    Open the coded letter, 01-letter-to-client-anonymise.docx, and search it (⌘F on a Mac, Ctrl+F on Windows) for Kettleby, the name in the page header: nothing should answer. Search for 900418, the end of the mobile number hidden in the comment: nothing should answer either. Then GB82, the start of the IBAN. Where they were, you now read codes such as NAME-001 or IBAN-001.

  6. Search the coded payroll

    Open the coded payroll in the same subfolder and search for Wrenfield: nothing. The names, National Insurance numbers, IBANs and dates of birth are codes. The salaries are still readable: once the people are codes, a salary no longer says whose it is. If the figures themselves must stay hidden, change the folder's Numbers setting before coding.

  7. Find what stays readable on purpose

    Open 03-services-agreement-anonymise.md with a text editor (TextEdit, Notepad). The parties and signatories are codes, but the fee, £48,500, is still there. This is the limit to know: an amount written in the body of a contract or a PDF stays readable, whatever the Numbers setting. If the amount itself is the secret, do not give that document to an AI as it is.

  8. Notice the file that was not coded

    The identity card scan, 04-id-card-scan.jpg, has no coded copy: a watched folder codes Word, Excel, CSV, text and PDF files that contain text, not images or scanned PDFs. A scan is read on your computer from Claude Desktop, with a proofreading step (lesson 3). Until then, it does not go to an AI.

What can go wrong

Check yourself

Three questions, answered on this page. Nothing is sent anywhere.

1. Which file do you give the AI?
Answer

The copy in the Hidden-Tokens subfolder whose name ends in -anonymise The originals keep every name by design. The copies ending in -anonymise are the ones to share; the NE-JAMAIS-PARTAGER file is the key and never leaves your computer.

2. In the coded contract, the fee of £48,500 is still readable. Why?
Answer

Amounts written in the body of a contract or a PDF stay readable by design The AI needs figures to work, so an amount in the text of a contract or a PDF stays readable whatever the Numbers setting. If the amount itself is the secret, don't give that document to an AI as it is.

3. The identity card scan has no coded copy. What do you do?
Answer

Read it from Claude Desktop, with local OCR and your proofreading (lesson 3) A watched folder codes Word, Excel, CSV, text and PDFs that contain text, not images or scanned PDFs. A scan goes through Claude Desktop, where it is read on your computer and you check the text before it is coded.

Frequently asked questions

Does Hidden Tokens change my original files?

No. The originals stay as they are; the coded copies are written next to them, in the Hidden-Tokens subfolder, and their names end in -anonymise. Those copies are the only files you give an AI. The same subfolder holds the key file, whose name says NE-JAMAIS-PARTAGER (never share): it must never go to an AI, an email or a shared drive.

Which files does a watched folder code?

Word (.docx), Excel (.xlsx, .xlsm, .xls), CSV, PDF files that contain text, and text files (.md, .txt). Images, scanned PDFs and PowerPoint files are not coded by the watched folder: scans go through Claude Desktop (local OCR, then your proofreading), and a file the folder skips should not be sent to an AI as it is.

Is the coded copy anonymous?

No, it is pseudonymised. The key on your computer turns the codes back into real values, so under the GDPR the data remains personal data for you. A combination of clues (a rare job title, a town, a year of birth) can also point to a person; Hidden Tokens flags such documents when you work from Claude Desktop.

Written for app 0.14.4 and connector 2.42.0 on . Not yet replayed on screen on Mac and Windows: if a label differs from what you see, trust your screen and tell us.