note 71075 deleted from function.mt-rand by cmb

Note Submitter: Chris ---- >Running the output of Mersenne Twister through an unkeyed >secure hash is NOT a good way to >make it secure, because it'll >still have a relatively small internal state which, if >recovered, >would allow reproduction of the keystream. A better idea >would be to encrypt the >output with a keyed encryption >algorithm - but if you were going to do that, you wouldn't >>need a psuedorandom number generator at all, because a >counter would be just as good. Not true. Mersenne Twister has an ENORMOUS amount of internal state - 4992 bits, bigger than practically any cipher's key length. The point of a secure random number generator is that you cannot predict future outputs based on past OUTPUTS, which is why a hash is applied. Clearly you can predict the future output of any pseudorandom number generator if you can acquire the internal state - a better algorithm will never solve this problem. If you use keyed encryption, recovering the key allows you to predict future outputs.