What it really means that data never leaves your server room

The phrase data sovereignty gets used very loosely. Here are the questions worth asking any AI vendor.

Many vendors promise data security while meaning encryption in transit to their own data centre. That is not the same as processing on your premises.

Control questions

  • Where does the model that produces the answer physically run?
  • Does the system need an internet connection to start at all?
  • Does the vendor collect telemetry, and if so, what exactly is in it?
  • Does the content of queries go anywhere outside your infrastructure, even briefly?
  • Can the whole deployment be rebuilt from scratch using resources you hold locally?

Three levels that are easy to confuse

ModelWhere the data sitsWhat constrains you
Vendor cloud serviceWith the vendor, often outside EuropeTrust in the contract and the vendor's policy
Private cloudIn a rented, logically separated environmentDependence on the cloud provider and its availability
Deployment on your premisesOn your hardware, in your networkYour hardware, your team, your procedures

What sovereignty does not solve

A system running on your premises does not automatically become compliant. You still need access control, an audit trail, a risk assessment and a decision about who is accountable for what the model produces. Sovereignty removes one risk - data leaving the building - and leaves everything else to design.

Accessibility

Text size
High contrast
Underline links
Pause animations