Document Anonymization for AI
Use ChatGPT, Claude or Gemini on real work without exposing personal data. We set up a gateway in your network that masks names, IDs and account numbers before the text goes out and restores them in the answer.
What the gateway does
Finds personal data
Names, ID and passport numbers, IBANs, card and account numbers, phones, emails, addresses, dates and API keys.
Your own patterns
Add internal identifiers such as client numbers, case references or project names.
Consistent placeholders
The same value always becomes the same placeholder, so the model can still tell people apart.
Privacy Mask engine
The same detection engine as our Privacy Mask extension, with checksum validation for cards, IBANs and national IDs.
Prompts, files and apps
Set up for chat prompts, uploaded documents or calls from your own applications, depending on how your team works.
Audit log
Can keep a record of what was masked and which model received the text, for your risk and compliance teams.
How it works
Detect
Rules with checksum validation, and a local model if you want one, find personal and confidential data.
Mask
Each value is replaced with a typed placeholder such as [PERSON_1] or [IBAN_1].
Process
The masked text goes to the AI model you use. It never sees the real values.
Restore
Placeholders in the answer are swapped back to the real values inside your network.
Works with the model you already use
The gateway sits in front of any model behind an API, including a model running in your own network.
Just need it in the browser?
Other ways to keep personal data out of public models.
Why teams mask instead of banning AI
Less exposure
Public models only ever receive placeholders, not the people and accounts behind them.
Central control
One set of rules for everyone, instead of hoping each person removes data by hand.
Same tools as today
People keep using the AI chat or model they already know; only the data that reaches it changes.
Who uses it
Teams that handle personal data every day.
Frequently asked questions
Have a question that is not here?
Which workflow should we start with?
Choose one route where staff send sensitive text to AI: chat prompts, uploaded documents or calls from an internal application. Define the data to mask and the model receiving the text before adding more routes.
Is it a ready product or a project?
A project. Detection uses the same engine as our Privacy Mask extension; the gateway, its connections to your tools and the logging are set up for your systems.
How do we include internal case or customer references?
Provide examples of the identifiers and the formats they follow. Custom patterns can be added for values such as case numbers or client references, then checked on sample material before rollout.
What should we check in a masking pilot?
Test the sensitive fields in your own material, repeated names and identifiers, and replies that use those values. Inspect the text sent to the model and the restored answer before choosing the workflow for rollout.
When is Privacy Mask enough for our needs?
Use Privacy Mask when people need to mask their own chat prompts or files in Chrome. Choose a company gateway when masking must be built into shared workflows, internal software or API calls.
Use AI without exposing client data
Tell us which models your team uses and what data it handles.