Glivent answers
How can Glivent use cloud AI without sharing sensitive information?
Direct Answer
A Glivent Hybrid workflow can keep sensitive information on the onsite Glive Pod and send only the material an approved cloud model needs. The workflow can remove names, contact details, account numbers or other identifying fields before the cloud step, then return the result to the controlled process for review.
How a Hybrid workflow works
The Glive Pod receives the source document or record inside the business. A local step identifies the fields that the cloud task does not need and removes or replaces them. The approved cloud model receives the smaller, redacted input and returns its result to the Glive System.
The rest of the workflow still runs through Glive Flow. Staff can review the output, compare it with the approved source and decide what happens next.
What can be removed before cloud processing
The exact fields depend on the task. They may include names, addresses, contact details, dates of birth, account numbers, patient identifiers or confidential commercial references. A workflow can also replace a name with a temporary reference so the result can be matched back to the correct record onsite.
Glivent sends the cloud model only the information required for its defined job. A model analysing document structure, for example, may not need to know who the document belongs to.
When redaction is not enough
Removing obvious fields does not always make information anonymous. A rare combination of details or the substance of a document may still identify a person or expose something confidential.
Glivent checks the full input and purpose before approving a cloud step. If the remaining material is still too sensitive, the task can stay on the Pod, use another method or remain with staff.
Why use a cloud model at all
Cloud models can be the better choice for some complex language, reasoning or document tasks. Hybrid AI lets the business use that capability without sending the complete source record by default.
The decision is based on the task, model performance, processing location, provider terms, cost and the consequences of an error. The approved data path is documented so the business knows what stays onsite and what leaves it.