Masking when you paste, attach or send, and keep context

You can always mask a highlight yourself, in the window on the highlight or in the side panel. On top of that, there are three moments where BeeSensible can do it for you: when you paste text, when you attach a file, and when you send. Keep context then puts the masked values back into the answer, on your screen only. Whether you see these four functions is up to your organisation. Whether they are on is up to you, unless your organisation has locked them.

#Where the switches are

Open the side panel and go to Settings. Under When to mask you find When pasting, When attaching and When sending; below that sits Keep context. The note above the switches says it too: you can always mask by hand.

Your organisation picks a state for each function. What that means for you:

For ChatGPT, Claude and Gemini, your organisation can deviate from its own state per tool. So a switch can be locked in one tool and free in another.

#Masking when you paste

Some AI tools already send along what is in the text field while you are still working on it. For pasted text, masking afterwards comes too late. That is why BeeSensible checks pasted text before it lands in the field.

#How it works

1. You paste text into ChatGPT, Claude, Gemini or Copilot (copilot.com).
2. Above the text field you see Masking what you paste, with what was found, for example "3 sensitive details found in what you paste", and a placeholder such as [NAME1] in each spot. Nothing is in the text field yet.
3. You choose Paste masked, Paste original or Cancel.

If nothing was found, the text is pasted as it is. If you do not want the preview any more, tick From now on paste masked right away; in the settings the same choice is called Paste masked right away, without the preview. Your organisation can make that choice for you as well: always with the preview, or always masked right away. In that last case there is no Paste original.

#Good to know

• Formatting is dropped: BeeSensible pastes plain text.
• Files and images on the clipboard take the AI tool's usual route.
• If detection gives no answer, you see "What you paste could not be checked" and decide yourself: Paste anyway or Cancel. BeeSensible never pastes unchecked text on its own.

#Masking when you attach

A file you add to a conversation in ChatGPT or Claude, through the upload button, by dragging or by pasting, is held back first. That gives you the chance to anonymise it before the AI tool sees it.

#How it works

1. You add a file in ChatGPT or Claude.
2. Above the input field you see File held by BeeSensible: the file has not been added to the conversation yet and has not been checked.
3. You choose Check and mask, Add the original or Cancel.

Choose Check and mask and the file opens in the Documents panel of the sidebar, with the detected values marked. After anonymising, you add the clean copy from the result card. See Redact a document before you share it.

#Good to know

• Needs the Documents module. If your organisation does not have it, this switch is not there.
• Held back is whatever BeeSensible can check: PDF, scans and images, Word, Excel, PowerPoint, OpenDocument, RTF, plain text, Markdown and CSV. Other file types pass unchecked. If your organisation switched the extra formats off, only PDFs are held.
• Add several files at once and they take their turn one by one: "After this, 2 more files."
• ChatGPT and Claude only.

#Masking when you send

Masking every value by hand takes time, and sometimes Enter comes too quickly. With this switch, masking happens the moment you send.

#How it works

1. You write your message in ChatGPT, Claude, Gemini or Copilot (copilot.com) and press Enter or the send button.
2. Everything BeeSensible still had highlighted becomes a placeholder.
3. Then your message goes through. In the sent message you can see what was masked.

Send while your last words are still being checked, and BeeSensible waits for that check, at most two and a half seconds. If the wait takes a little longer, "BeeSensible is checking your message" appears above the field. If the result does not come in time, the message goes through as it stands.

#Good to know

• A language, such as "English", is never masked automatically: in a prompt that is almost always the instruction. Masking it by hand still works.
• It remains an aid. BeeSensible can miss something, so read your prompt before you send it.

#Keep context

Sensitive values are often exactly the context an AI tool needs. Mask them and the answer comes back full of placeholders. With keep context, BeeSensible puts the real values back on your screen. The AI tool only ever sees the placeholders.

#How it works

1. You mask something in ChatGPT, Claude, Gemini or Copilot (copilot.com): by hand, when pasting or when sending. A name becomes, say, [NAME1-K7]; the short tag after the number belongs to this conversation.
2. BeeSensible remembers on your device which placeholder stands for which value.
3. In the answer you read the real name, with a dotted line under it. That is how you see what the AI tool did not see.

The same name keeps the same placeholder throughout a conversation, also in a follow-up question. A new conversation starts empty. Under the switch, Works in lets you turn it on or off per AI tool.

#With documents

Anonymise a document in an AI conversation and you choose per document: Black box or Placeholder. With placeholders, the document belongs to the conversation: a name that appears in both the document and your prompt gets one placeholder, and the answer about the document reads with the real values just the same.

#Good to know

• The memory lives on your device only and never reaches BeeSensible. By default it lives in the browser session: it is gone when you close the browser, sign out, or after a day. If your organisation chose 30 days, it is stored encrypted on your device and you can still read last week's conversation with the real values. It then expires after 30 days without use, and when you sign out.
• Copy from the answer, with a selection or with the AI tool's copy button, and you get the real values. Share or export the conversation from the AI tool, and it contains placeholders.
• The AI cannot do arithmetic with a placeholder. Ask "how old is [AGE1] in three years" and you get [AGE1] + 3. For a national ID, IBAN, email or phone number, masking costs nothing; for an age, amount, date or place, the answer can become less useful.
• The AI does not know who [NAME1] is and avoids "he" and "she". An answer can read stiffer because of it.
• When asked to translate, Copilot sometimes translates the placeholder itself. BeeSensible does not restore that one; in that spot you read the placeholder.
• Does not work in Microsoft 365 Copilot or in the desktop app.
• Your organisation's figures only show that masking happened, and whether it was by hand or when sending. Never what.

#What your admin sets

The state per function, the flow for pasting, who keep context applies to and how long its memory lasts, and a deviation per tool. All of that is in Masking: paste, attach, send and keep context for admins. Names that only mean something inside your organisation, such as projects and clients, get the same highlight and a placeholder with their category in it, such as [ORG-PROJECT1]; how your admin keeps that list is in Custom terms.