Protect sensitive data while your team works with AI.
BeeSensible highlights details such as a customer name, ID number or account number in the text. People can mask them before they share anything, in the tools they already use. The help arrives exactly when they apply what they learned in training.
ChatGPTThe marking in ChatGPT. Demonstration with fictional data; click a marking.
The AI tool gets the question, not the person.
What the employee types
Write a reply to Ingrid Hansen, SSN 078-05-1120. The refund to card 4532 1234 5678 9010 is on its way.
What the AI tool receives
Write a reply to [NAME1], SSN [SSN1]. The refund to card [CARD1] is on its way.
Product example with fictional data, not a test result.
Up to 95% better protection than without BeeSensible.
People no longer have to strip sensitive details out of their text themselves. BeeSensible finds them: names, ID numbers, account numbers, addresses and medical details, in Dutch and English. One click and they are masked, or it happens automatically on paste or send. Your team makes full use of AI, without customer data slipping through unnoticed.
How we measure this
In our own detection test, Bombus recognised 95 of every 100 sensitive values (F1 score 0.96), across all 65 data types, in Dutch and English, on a frozen test set not used for tuning. Without BeeSensible nothing is recognised automatically. The figure says what the detection finds, not how many data breaches are avoided. How we test the detection
Help in the text, while people work.
Someone pastes a customer email into ChatGPT to turn it into a polite reply. The ID number and the account number are easy to miss.
BeeSensible recognises 65 kinds of sensitive data and your own terms, such as project and client names, as people type or paste: yellow for sensitive, red for critical. People can mask a detail with a placeholder such as [SSN1], remove it, or leave it. This works in AI tools such as ChatGPT, Claude, Gemini and Copilot, and in email and chat.
The AI tool gets what it needs to answer, without the person behind it. You decide how warnings and masking are set up; the options are described in the documentation.
When pasting
Pasted text is checked before it lands in the field. People see what was found and paste it masked, paste the original, or cancel.
When sending
If something is still highlighted when someone presses Enter, it can become a placeholder on the way out.
Keep context
The AI tool only sees placeholders. The real details come back in the answer, on the employee's own screen.
Custom terms
Names that only mean something inside your organisation, such as projects and clients, are highlighted like any other sensitive detail.
Files leave clean, to an AI tool or out the door.
Not everything sensitive is typed. An intake report, a CV or a contract is added to a conversation as a file, or forwarded.
Open the file in the dashboard or the extension: PDF, scanned too, Word, Excel, PowerPoint, images and text. Everything found is ticked; the employee reviews, draws a box where needed and exports a clean copy in the same format. Comments, tracked changes and metadata are checked and cleaned too. With mask on attach switched on, a file you add in ChatGPT or Claude is held first: check and mask, add the original, or cancel.
The document that leaves contains only what belongs in it. After redacting, BeeSensible reads the result back and warns if something is still there. In the desktop app this works for PDF.
Documents in the dashboard. Demonstration with fictional data.
A file held in ChatGPT. Demonstration.
Anyone opening an AI tool sees what the organisation decided.
An agreement on the intranet is not always to hand in the middle of the work.
When someone opens a tool you have decided about, that decision shows in the browser. For DeepSeek: not allowed, with the button to Microsoft 365 Copilot first. For a tool with limited use: the condition in your own words. For a tool without a decision: a quiet note, if you want one.
The agreement reaches people when it matters, with the alternative one click away. Whether someone can carry on after that is your choice. BeeSensible does not block anything on its own.
DeepSeek is not allowed here
Your organisation does not allow this tool for work. Use an allowed alternative.
Try the three notices. Demonstration with fictional data.
Building on training
What people learned in training comes back in their work.
Training lays the groundwork: what counts as sensitive, which tools are allowed and why. BeeSensible brings that knowledge back at the moment people use it, in the text they are working on. Employees get an explanation and short videos on what they see and what they can do.
In the text itself
The highlight appears where people work: in the AI tool, the email or the chat.
People act for themselves
Mask, remove or leave it, within your organisation's settings. A highlight is a reminder, not a ban.
See what lands
How often sensitive details are adjusted before sending and what happens after a notice, in totals. Useful for tuning your training.
Trust and getting started
Safe for your data, and quick to start.
What happens to your data
- Text is checked on our own servers at Scaleway and Hetzner, two European companies, and deleted straight away.
- The dashboard shows totals for the whole organisation, never per employee. Figures appear from five active users.
- You get a processing agreement. The product DPIA and security information are available on request.
How to start
- Try BeeSensible free for 14 days, with every module and no credit card.
- Install the extension on a few workstations, or have IT roll it out centrally through Intune or Group Policy. That works on every plan.
- Employees get an explanation and short videos on what they see and what they can do.
Try it on your own AI tools.
Your team just keeps working and sees the marking from day one. Fourteen days free, no credit card. Rather see it work in your tools first? Ask for a demo.
14 days free, no credit card. Rolled out across your organisation within an hour.
