From: Matt Blaze <mab@crypto.com>
Date: Wed, 1 Feb 95 13:37:44 PST
To: norm@netcom.com (Norman Hardy)
Subject: Re: ESP Unix encrypted session protocol software
In-Reply-To: <ab559e8c000210040b91@DialupEudora>
Message-ID: <199502012138.QAA26261@crypto.com>
MIME-Version: 1.0
Content-Type: text/plain



>I trust that that the attack refered to is the "man-in-the-middle". I find
>it very curious that there is a simple fix to the attack for the enctrypted
>voice channel. Each unit displays to its human a few bits of g^(xy). One
>human quotes them vocally to the other. If there is a man in the middle the
>bits are unlikely to match. What I find curious is that there seems to be
>no automated analog to this precaution. It has to do with the difficulty of
>substituting the vocal signals that code these bits. This is too hard for
>either computer or man (in the middle). I write to stimulate a solution. I
>have none.
>
>
The reason there's no "computer" analog to the "anti-spoofing vector"
for human-human voice communication lies in the definition of
authentication.  In a formal sense authentication here means binding a
secret that only you know to the encrypted channel.  In the case of voice
communication over an encrypted link, that "secret" consists of the ability
to hold a convincing exchange that sounds like your voice.  You bind the
secret to the channel by speaking a hash of the key.  Computers, not
pre-equipped with biological mechanisms for establishing who they are,
need to use another secret (like knowledge of the secret part of a public
key signature pair) to which only the computer you want to authenticate has
access.

The encrypted human voice authentication scheme is only as strong as it
is hard to spoof voices.  Digital signature authentication is only as
strong as it is hard to break the signature scheme or compromise the
signing key.

-matt