Security mode

Security mode is a switch in the chat for the most sensitive conversations, such as those about client mandates or personnel data. When it is on, a model your admins chose for the purpose does the work, and every detected detail is pseudonymised, with no way to send any of them as typed.

You only see the switch if your admins offer security mode and have chosen a model for it. Setup is described in Setting up security mode.

Turning security mode on

Open a new chat

Click New chat in the sidebar. Security mode always applies to a whole chat, so switch it on before the first message.

Click the lock

In the input field, click the lock next to + (Turn security mode on). The lock is highlighted and now reads Secure.

Check the settings

Below the input field there is now a line that starts with Security mode:, names the model and continues with "details are pseudonymised, no web search". The model picker is locked and shows Fixed by security mode. Web search is hidden.

Write and send

Work as in any other chat. If your workspace uses the Pseudonymise protection mode, the review panel shows what will be replaced. How that works is described in Protecting personal data in chat.

In the input field you see the lock labelled Secure, and below it the security mode line. In the sidebar the chat looks like any other.

What security mode guarantees

In the settings, your admins see these commitments under What security mode guarantees:

Only the chosen model. The picker is locked.
Pseudonymisation is always on, even if the workspace is otherwise set to Off.
No web search.
Send original is disabled. No single detection can be waived.

Compared with an ordinary chat, this means:

	Ordinary chat	Security mode
Model	free choice or Auto	fixed
Pseudonymisation	depends on the workspace's protection mode	always, with every category
Sending individual values unchanged	possible in Pseudonymise mode	not possible
Web search	depends on the workspace	off
Content from Company Brain	depends on the brain's level	always pseudonymised
Outlook, Google Drive, Microsoft 365 used by the model	possible when connected	not available
Attaching images	depends on the protection mode	not possible
Creating images	depends on the workspace	not available

You can still attach files such as PDF, Word or Excel. Their content is pseudonymised before it reaches the model.

Turning the mode off again

As long as a chat contains no messages, one click on Secure switches the mode off again. Once the first message has been sent, that is no longer possible. The hint begins with Cannot be switched off in a chat that has started. In that case, start a new chat.

A new chat in security mode: the message box with “Secure” and the line naming the fixed model, pseudonymisation and no web search.

If your admins have set New chats start in security mode, the lock is already active in every new chat. You can still switch it off before the first message.

You can switch the mode on at any time, even in a chat that has already started. The earlier messages, however, were sent under the ordinary rules. For confidential topics, start a new chat in security mode from the outset.

When security mode is worth it

Client data: Lea Keller at Kieselmatt Treuhand AG prepares the annual accounts for Holzwerk Brunner GmbH and discusses payroll lists and bank statements.
Personnel data: references, salary questions, employment contracts, health information.
Contracts and negotiations where nobody should send a name in plain text by accident.

You do not need it for general questions, web research or images. An ordinary chat is more practical there.

When the credits are used up

If the workspace's credits for this month are used up, the economy model that then applies to everyone also answers in security mode. Pseudonymisation, the region setting and the locked model picker all stay in place. Both notices then appear below the input field. More in Choosing a model.