Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10 5/3/83; site cbosgd.UUCP
Path: utzoo!watmath!clyde!floyd!harpo!ihnp4!cbosgd!mark
From: mark@cbosgd.UUCP (Mark Horton)
Newsgroups: net.crypt
Subject: Re: One-time pads
Message-ID: <986@cbosgd.UUCP>
Date: Sun, 19-Feb-84 19:15:31 EST
Article-I.D.: cbosgd.986
Posted: Sun Feb 19 19:15:31 1984
Date-Received: Mon, 20-Feb-84 08:59:40 EST
References: <2540@azure.UUCP>
Organization: AT&T Bell Laboratories, Columbus
Lines: 20

I've seen two people claim that the chances of the cleartext getting
sent out unencrypted due to a long string of zeros in the key is no
greater than some other message coming out due to a random key.

I think these people are assuming that the random number generator
used is sure to generate uniformly distributed random numbers.  With
a good psuedo random number generator, you can easily prove that this
is the case.  But with a true random number generator, based on noise
from some analog source, for example, you are NOT assured that the
numbers are uniform.  Maybe there were excessive sunspots the day
the key was generated, and the noise levels were very low, resulting
in all the analog values being sampled being below the zero cutoff,
and a zero random number value for each try.  Or perhaps one of the
components breaks, with the same result.

The obvious thing to do is to have the algorithm watch the key, and
if it sees a long string of zeros or ones, to refuse to send the
message, but to alert an operator instead.

	Mark Horton