You'd have to correct for the biases. Not only in probability, but subsequent calls are also not independent.
Re:/dev/random – a new approach
by thaumasiotes on Sunday April 24, 2016 @09:38PM
(#11562148)
Sure, I realize that. (Although "subsequent calls are not independent" means it's not actually refusing to respond "at random".) I'm just pointing out in a facetious, emoticon-marked manner that "it behaves randomly" is kind of an odd insult to level against a source of randomness.
I think the point is that it should behave predictably (but of course give you random data). Behavior is a "larger" concept than just what kind of data you get from a thing.
Why would this be a problem? Read one byte from /dev/random, and observe one of 257 possible values chosen at random. :D