From: stewarts@ix.netcom.com
Date: Wed, 4 Sep 1996 11:02:13 +0800
To: paul@fatmans.demon.co.uk
Subject: Re: Secure anonymouse server protocol: comments please
Message-ID: <9609032156.AA05838@anchor.ho.att.com>
MIME-Version: 1.0
Content-Type: text/plain


At 07:24 PM 9/2/96 +0000, paul@fatmans.demon.co.uk wrote:
>The following is a very sketchy plan for a secure protocol for an 
>anonymous server which allows replies without storing a recipient 
>database in the clear.

Several people have talked about this sort of thing recently,
inluding William Geiger, Doug Floyd, and myself.
Lutz Donnerhacke's Jenaer Anonymous Service <anon@as-node.jena.thur.de>
actually implements it (send it mail saying "help".)
Rather than using a human-selected userid, it uses the
PGP keyid to make IDs like anon-1a2b3c4d@as-node.jena.thur.de.

>This system has 1 huge fault, we can encrypt a uses ID with the 
>servers public key to see what his ID in the encrypted database is 
>and therefore identify him, maybe we need two seperate server public 
>keys, and when IDs come in encrypted with key1 (the one it releases) 
>it decrypts with secretkey1 then encrypts with publickey2 (the one it 
>keeps secret)

If you encrypt the id using raw RSA and constant padding, this is a risk.  
If you encrypt it using PGP, which uses a random session key, it's not.
If you encrypt it using raw RSA and pad the id with a random nonce, it's
also no risk.  In the latter two cases, the encrypted material is different
every time, so you can't compare with previous messages.

The Jenaer nymserver avoids this by using a remailer approach -
you send an encrypted message with a Reply-To: header telling where
to send the accumulated mail (which may, of course, be another nymserver),
and it delivers it using mixmaster.  This frees you to send your pickup 
requests by anonymous remailer as well.  It's still not risk-free,
since if Bad Guys crack the remailer or force the operator to operate it
while they monitor it, they can see pickup requests, but it's far
more difficult to do that than to just steal the box, and there's
no database on the box that's useful to steal.  Lutz does recommend
chaining your Reply-To: to another nymserver, but it's already very secure.

I don't remember if he gets fancy and requires the pickup requests
to be signed by the key of the owner or not; the only difficulty with this
is the syntax of PGP, which is "fixed in 3.0".

Hal Finney has also suggested a system that, instead of delivering
anonymous email to the recipient, sends a message saying
"You have anonymous mail, receipt #123456.  Send back this ticket
to pick it up."  and you can extend the syntax to handle
automatic blocking requests and automatic deliver-everything requests.
This is fairly easy to extend for anonymous mailboxes
and datahaven code.  I've wavered between the delete-on-retrieval model,
which is fine for email and not very useful for samizdat,
or the delete-after-some-time-period-or-request model, which is useful 
for both but makes it easy for users to turn you into the local
pirate-warez-and-child-pornography server.  If you extend the model
and charge digicash for storage, it becomes a much cleaner solution.

#			Thanks;  Bill
# Bill Stewart, +1-415-442-2215 stewarts@ix.netcom.com
# <A HREF="http://idiom.com/~wcs"> 	
# You can get PGP software outside the US at ftp.ox.ac.uk/pub/crypto