Path: cactus.org!news.dell.com!swrinde!howland.reston.ans.net!math.ohio-state.
+     edu!caen!msuinfo!netnews.upenn.edu!netaxs.com!unix3.netaxs.com!soneill
From: soneill@unix3.netaxs.com (Steve O'Neill)
Newsgroups: sci.crypt

Subject: Re: Doing Better than XOR in RC4-like Algorithms
Date: 16 Nov 1994 12:31:51 GMT
Organization: Netaxs Internet BBS and Shell Accounts
Lines: 24
Message-ID: <3acu3n$36j@netaxs.com>
References:  <3a7llc$kej@netaxs.com> 
NNTP-Posting-Host: unix3.netaxs.com

In reply to Stewart Strait's answer to my statement that using the same key
for two messages in RC4 doesn't necessarily make analyzing the resulting
cryptexts trivial, I must state that his comments on finding frequencies of
occurence are quite valid _if_ the plaintext is English text. However, in my
original post, I did say that "in the general case", XORing out the common
key doesn't help. To me, this means that the "plaintext" can be binary,
compressed text, Fijian text, or anything else. If an attacker doesn't know
in advance what the nature of the original messages is, having the XOR of
two messages in front of him doesn't give him _any_ information. If it did,
then every one-time pad ever devised would have been broken, since XORing
one unknown message with another unknown message is equivalent to a one-time
pad.

Of course, this illustrates the old dictum that the biggest difficulties in
keeping information secret are operational, not functional. IF you can keep
possible attackers from gaining _any_ knowledge of the nature of your
transmissions, then an encryption algorithm like RC4 can keep your
information confidential. In the real world, however, this is almost
impossible to guarantee. So, from an operational point of view, changing
keys for every transmission is an absolute requirement.

Steve O'Neill