Docx Anonymizer recognizes 50 data categories and supports nine formats

Sending a contract or spreadsheet to an AI chat is easy. Making sure that passport data, colleagues’ comments, a hidden Excel sheet, or the author’s name in the metadata did not travel with the main text is harder. This is the mundane but unpleasant problem addressed by the Docx Anonymizer Chrome extension: it searches for personal data before a file is uploaded to an external service and lets users remove traces they would not normally see.

In the new version, the developer expanded support beyond Word documents. The tool works with Microsoft Office spreadsheets and presentations, as well as TXT, Markdown, CSV, JSON, and log files. The number of recognized categories has increased from 21 to 50. For a developer or product team, the rationale is straightforward: before giving an AI system a customer export, meeting minutes, or a technical log, the file can be checked with the same tool.

How it works in practice

Consider a spreadsheet of support requests. The visible cells contain a customer’s name, phone number, and problem description. A hidden sheet still contains an internal sample, a manager has entered a contract number in a comment, and a formula references a cell containing an email address. If someone simply copies the main sheet or replaces a few obvious values, some of the data will remain inside the file.

Docx Anonymizer searches for more than names and contact details. Its stated coverage includes details from Russian domestic and international passports, driver’s licenses, military IDs, residence permits, migration documents, compulsory health insurance policies, vehicle registration certificates, and vehicle passports. VINs, government-issued and cadastral numbers, IP and MAC addresses, health information, and other sensitive categories are recognized separately. Passport data is now parsed into four distinct fields: series and number, issuing authority, issue date, and department code.

For text-based formats, the replacement scope has an important restriction: CSV delimiters and JSON keys are not changed, so the data structure should be preserved and the JSON should remain valid. This is useful for test datasets, logs, and exports that need to be loaded back into an application or processed by a script after anonymization.

Hidden data is a separate risk

The extension displays comments, revision history, hidden text, hidden sheets and slides, as well as cached formulas and charts. These elements can be removed before the document is sent. This check is especially important for Office files: a document that looks clean may not be clean inside the archive that technically makes up a DOCX, XLSX, or PPTX file.

Automation, however, does not guarantee that leaks are impossible. Heuristics may miss an identifier written in an unusual format or, conversely, mistake an ordinary number for personal data. The developer also warns that replacing values inside formulas can break some formulas. The result therefore needs to be checked, while the original file should be stored separately.

What teams should do

The tool should be treated as an additional barrier before documents are shared with AI systems, not as a complete replacement for a data-protection process. First, determine which files may be uploaded to external models at all. Then remove unnecessary sheets, comments, and revision history, run anonymization, and inspect the resulting file manually. For routine use, it is helpful to maintain a control set of documents containing known data types and test every extension update against it.

The project’s main value lies not in the number of supported formats, but in its reminder of the true attack surface for data leaks. Personal data does not exist only in paragraphs. It hides in formulas, metadata, and internal document layers—and those are precisely what people most often forget to inspect before talking to an AI system.