Data protection in a brain
Every brain has its own data protection level (Open, Pseudonymise or Strict), independent of the workspace's protection level.
In a chat, an IBAN is often a mistake; in a brain it can be the answer. That is why every brain decides for itself what a model gets to see from its documents.

The three levels
| Open | Pseudonymise | Strict | |
|---|---|---|---|
| Passages in chat | as stored | with placeholders, put back in the answer | like Pseudonymise |
| Title, summary, overview | from the original text | from the pseudonymised text | from the pseudonymised text |
| When adding documents | everything is accepted | everything is accepted | documents with an AHV, tax ID or social security number, IBAN, card number or UID/VAT number are refused |
| Images and scans | are read | are not read | are not read |
Pseudonymise replaces whatever the workspace's protection level pseudonymises (Data protection). With the workspace on Off or Standard, these are the six hard identifiers; names stay readable. Only with Strict, or a matching choice under Custom, are names, addresses and contact details replaced too.
What always applies
- The real text is stored. Anyone with access sees the original values in the preview.
- The search index uses the original values, so a question about a name finds something. A model at AWS in Zurich does this; it produces search values only, no text.
- Workspace on Strict: nobody can set a brain to Open. A new brain starts on Strict there.
- Security mode: there too, passages from a brain follow the brain's own level (Security mode).
- Edit in chat: a document you edit in chat is a file of that chat there; the workspace's protection level applies.
No detection is perfect. How pseudonymisation works explains the method and its limits.
Changing the level
- When you switch to Strict, Custodos checks the existing documents. If some contain hard identifiers, the dialog "Documents with sensitive identifiers in this brain" names them and you choose Remove and switch or Keep anyway and switch. Kept documents still reach the model.
- For chats, the new level applies from the next message. Every change is recorded in the activity log.
- After switching away from Open, images and scans not yet read are no longer read.
Which level fits
| Brain | Level |
|---|---|
| Handbook with internal rules, scans and images | Open |
| A client mandate's correspondence with account and AHV numbers | Pseudonymise |
| Templates that should never contain account or AHV numbers | Strict |
Frequently asked questions
The model saw placeholders. Custodos puts the real values back into the answer for you. That is exactly how the level is meant to work.
Image content cannot be pseudonymised, so only brains set to Open accept images. Put the image into an open brain or transfer its content with Paste text.