Take a look at the documentation
Several methods are supported but the default uses cellular automata. It used to be rule 30, see Stephen Wolfram's "A New Kind of Science"
which has passed all of the tests for randomness. The order of its period is large, on the order of all possible states. For example, if the initial condition had 100 bits, the possible states are 2^100. (Of course there is more to randomness than the order of the period.) I don't know what rule is used now, but it performs even better.
At the Wolfram Science Summer School, we've had a few students investigate the randomness in cellular automata randomness generators. Someone could be interested in randomness as a matter of science, that would be for the Wolfram Science Summer School, but maybe someone is interested in randomness as a matter of technology (e.g. cryptography, signals, etc.) and that would be for the new program focusing on technological innovation.