Control the data. Govern the conversation.
Confidential AI governance starts with the data. Sovrata brings classification, encryption, access decisions and audit evidence together, so your organisation’s policies govern both the files people use and the AI acting for them.
If you are not cleared to open a document, neither is your AI assistant.
An AI assistant acts for a person and sees only what that person is cleared to see. The person’s permissions and the document’s classification determine access throughout the interaction.
Illustrative access comparison
The same question. Different clearance.
With clearance
The person can open the document. Their AI can use it to answer.
Without clearance
The person and their AI receive a response indistinguishable from the document not existing.
Both access decisions are logged for authorised review.
One policy engine. Every read.
One policy engine applies your organisation’s rules to every read: a file opened, a query run or an AI search. People keep their different roles, clearances and permissions, and AI acts within those same boundaries.
Classification, roles, attributes and object permissions all count before anything is returned. Rules apply at file, folder and column level, with row rules in Sovrata’s own SQL service, so the same question can get two different, correct answers.
With content routing enabled, governed chats use models approved for their security level. A conversation keeps its highest established classification, even if the next question is harmless. Your organisation decides which models may handle each level.
Document records connect supported conversations to source versions and their first and latest delivery. Separate inference audit events record the model, request classification and policy decision. Review the evidence and supported scope.