Custom terms

Bombus recognises a passport number or a name, but not that "Project Havik" is your acquisition or that "Atlas" is your client system. For those, your organisation keeps its own list. When someone types or pastes such a term into an AI tool, it gets a marking like any other data type, and after masking a placeholder that carries its category, such as [ORG-PROJECT1].

The list lives under Policy > Detection profiles, tab Custom terms.

#Adding terms

One at a time: fill in Term, pick a Category and click Add. Or in bulk with Import CSV: one term per line, optionally with a second column for the category, such as Project Havik,Project. Terms already on the list are skipped, and the screen previews how the first lines will land. Download CSV gives you the whole list back.

A term has 3 to 100 characters. The list holds at most 5,000 terms.

#Eight categories

The category decides the placeholder after masking. The ORG prefix shows, inside the prompt, that the value came from your organisation's list.

#Whether a category counts, and how strictly

That is not set on the term but in the detection profile of the app: next to Sensitive data, every profile has a Custom terms tab with, per category, whether it counts and at which level, Standard or Critical. The marking follows that level, yellow or red, exactly like any other data type; custom terms have no colour of their own. A category you never touched counts at Standard. See Detection profiles.

In the extension's list of found values, such a marking is labelled as coming from your organisation.

#How matching works

Literally, so you can predict what a term will mark:

• Capitals, accents, spaces and hyphens make no difference: Project Havik also finds project-havik.
• A term entered entirely in capitals only matches capitals. The abbreviation KIT leaves the word "kit" alone.
• Typos and compounds are not found: Havik does not mark Havikproject. If you want that, add the variant.
• When two terms overlap, the longer one wins.

#Where the list stays

• The extension searches on the device, after the detection result comes back. The list never goes to the model.
• The list is stored encrypted in BeeSensible's database and encrypted on the device. Signing out clears it from the extension.
• The figures only ever hold the category, never the term: under Sensitive data you see "Project (custom term)" and can filter on all custom terms. The audit log records that something was added or removed, and how many, not what.
• A change reaches every device within five minutes.

#In documents too

Custom terms also work when anonymising documents, with the profile that applies to Documents. In the desktop app that happens on the device itself, in a local scan.