Provider Confidence

Deutsch: Vertrauen in Anbieter

Not every provider deserves the same trust. Some process your data in your country under strict data protection laws, others elsewhere; a model on your own computer keeps everything with you. With the provider confidence, you rate your providers, see the rating while you work, and keep sensitive work away from providers you trust less.

You find these settings in "Settings" → "Provider Confidence".

The levels

From lowest to highest: "Untrusted", "Unknown", "Very Low", "Low", "Moderate", "Medium", "High".

In the predefined schemes below, providers that run on your own computer or server ("Self-hosted") always have the level "High". A provider that your organization prohibited is always "Untrusted".

Choosing a scheme

Instead of rating every provider yourself, you choose a scheme under "Select a confidence scheme":

Scheme	What it does
Trust all LLM providers	Every cloud provider gets "Medium". This is the default.
Trust LLM providers from the USA and Europe	Providers from other regions get "Low".
Trust LLM providers from the USA	Providers from other regions get "Low".
Trust LLM providers from Europe	Providers from other regions get "Low".
Trust LLM providers from Asia	Providers from other regions get "Low".
Trust only local LLM providers	Every cloud provider gets "Very Low".
Configure your own confidence scheme	You set the level of each service yourself.

The level applies to all your providers of the same service, whether you use them for chats, transcriptions, or embeddings.

Your own scheme

With "Configure your own confidence scheme", a table lists every service with a "Description": where it processes your data and what it promises, for example, that it doesn't use your data for training. Click the button in the "Confidence Level" column to set the level. A service you haven't rated yet counts as "Unknown", and that includes "Self-hosted".

Seeing the level while you work

When "Show provider's confidence level?" is on, which is the default, a shield next to the provider selection shows the level of the selected provider in a color, from red for low to green for high. You see it in the chat and in every assistant. Click the shield to open the "Confidence Card", with the description of the provider, its sources, its region, and its level.

The scheme and your own levels only appear in the settings while this switch is on. They apply either way.

Requiring a minimum

A minimum keeps providers below a certain level away from your work. They then no longer appear in the provider selection. You can set a minimum in several places:

• For all of AI Studio: switch on "Do you want to enforce an global minimum confidence level?" and choose the level under "Select a minimum confidence level".
• For one assistant: open the options of the assistant with the gear icon, switch on the preselection, for example, "Preselect translator options?", and choose "Select a minimum confidence level". The minimum only applies while the preselection is on. The Document Analysis assistant takes its minimum from each policy instead.
• For a tool: in "Settings" → "Tool Settings", column "Minimum provider confidence". A tool then only runs with providers at this level or higher.
• For a data source (beta): "Required provider confidence level" in the settings of the data source. Only providers at this level or higher see its content.

When the global minimum and the one of an assistant differ, the higher one counts. When no provider meets the minimum, the provider selection says so, and you need to set up a provider with a higher level in the settings.

When your organization manages AI Studio

Your IT department may choose the scheme for you, set a minimum, or prohibit providers. You then can't change these settings, and a lock shows why.

Where to go next

• Providers and Models: setting up providers and choosing which one is preselected.