If when one works with data in a dataset, we are working with a table in memory, and not the database itself. How do we enforce user concurrency controls?

Lets say two users get the same record loaded into a dataset with an adapter or command object. One performs a modifications and saves it, then the other one saves again, therefore overwriting the changes the first person made.

I haven't seen anything to deal with this, im pretty sure there is something done implicitly, and if not something has to be available.