rng, anyone?

[Matthew J Ghio]

hughes at ah.com (Eric Hughes) wrote:

> There is a problem with generating random numbers by repeated
> iterations of a hash function when these numbers will be used to
> simulate an encrypted message body.  The body can be seen to be
> generated by the algorithm.  All you do is to apply MD5 to the first
> block and see if it's equal to the second block.  This completely
> identifies the message as a hash-chain generation, and thus as a fake
> message.
> 
> Indistinguishability is a harder criterion to simulate than other
> notions of randomness.

Try xoring the output with a secret value between MD5 hashes.