AI Vocabulary

Segregation: is your data separated, or merely filed apart?

Logical segregation separates data by rules within shared infrastructure. Dedicated infrastructure separates it physically, on reserved machines. The first is the norm and works well; the second costs considerably more and does not address the same risks.

What logical segregation does

Your data and other clients' coexist in the same infrastructure, with an identifier distinguishing them. Every request is filtered to return only what belongs to you.

This is the model of nearly every professional service, including those used by banks and insurers. Properly implemented it is robust: separation is applied at every access, without exception.

Its weakness is that it rests on software. A configuration error can expose one client's data to another, and such incidents do occur.

What dedicated infrastructure adds

Machines processing only your data. There is no longer a neighbour, therefore no risk that a misapplied rule lets two clients meet.

It costs a great deal, it slows updates, and it removes none of the other risks: the provider still has access, the location question remains open, so does the question of which model is called.

The segregation least discussed

There is a third separation, inside your own organisation, and it is often the one that poses the real problem.

A firm maintains barriers between its teams: for professional-conduct reasons, to prevent conflicts of interest, because certain practices are covered by heightened confidentiality. A system giving the whole firm a view of every practice erases those barriers, without any data having left anywhere.

That is the question to ask first, before the hosting question: can the system reproduce the barriers that exist here?

It has a decisive practical consequence: a system unable to reproduce them forces its use to be confined to the least sensitive practices, that is to say those where it helps least. Internal segregation therefore decides the scope of deployment long before it decides the level of security.

What it does not solve

Neither model protects against a legitimate access wrongly granted. If a colleague receives rights they should not have, segregation works perfectly and gives them access to what was opened to them.

And dedicated infrastructure does not improve what concerns you most: the provider retains its administrative access, and the model called remains the same.

One further point: these barriers move. A team taking on a matter sometimes creates a new one for the duration of a dispute; a practice classified as sensitive may cease to be. An arrangement allowing barriers to be defined only at installation is therefore ill-suited to a firm, whose internal frontiers shift several times a year.

The answer to that test also determines what you can tell a client who asks who, inside your firm, has been able to see their matter.

Why it matters to a lawyer

Because dedicated infrastructure is often offered as the premium answer to a confidentiality concern, when it addresses a secondary risk at a high price.

The test that really informs takes one manipulation: ask what happens when a colleague queries the system about a practice they do not have access to. Do they get a refusal, silence, or a partial answer? And does the attempt leave a record?