Re: Hardening lists against spam attacks

[Hal Finney <hal@rain.org>]

I think this is an interesting theoretical discussion, although it's
not clear whether it is actually a good idea to try implementing this.

From: nobody@replay.com (Anonymous)
> A very good scheme, but why not give each subscriber a token when s/he
> subscribes? Something along the lines of:
>
> - - - - - - - - - - - - - - - - - - - -
> Welcome to Cypherpunks.
>
> Your unique token is: 0A553FC1771623109504522E31C07F44
>
> This token must appear either as the first line of the message body or in an
> X-Token: header for any mail you send to the list. Any messages sent to the
> list
> address without this information will be discarded.
> [...]
> Bob wants to post something anonymously. His token isn't associated with his
> user ID -- the only thing Majordomo knows about it is that it's in the token
> file and it's flagged as active. He sends the message through the remailer
> network with his token in it, and Majordomo validates it, strips it out, and
> passes the message to the subscribers, decrementing the number of messages Bob
> has remaining for that day.

This requires Bob to trust the server to keep his identity secret.
Although you _say_ that majordomo didn't associate the token with
the userid, how does Bob know that?  Certainly majordomo did, when
Bob subscribed, see the association between the userid and the token.
Now he has to trust that it has been forgotten.  Even if it has, what
about eavesdroppers on the list channel?  What about the operator on the
machine, who is peeking at what majordomo is doing?  This mechanism will
not provide enough anonymity for most posters.

An alternative similar to what I proposed earlier is for majordomo to
provide a blinded token, one which it doesn't see.  This would be used
specifically for anonymous postings.  It does have the problem that it
allows linking postings by the same pseudonymous nym - all will have the
same token.  But maybe we want to encourage that.

(The full proposal I made involved use-once tokens, just like online
digital cash, so that there would be no linkage and it would allow
real anonymity.)

> Mallory wants to spam the list. He subscribes and gets a token, which he
> uses to
> forward commercial announcements to the list. The list manager checks the logs
> to see which token was used, and reduces its posting limit or invalidates it.
> Mallory is no longer allowed to post, unless his token is reinstated (or he
> unsubscribes and resubscribes).

This unsubscribe/resubscribe issue has been mentioned before as a problem.
I am not too concerned with it, for a few reasons.  First, it may not be
too difficult to recognize that it is happening.  If the same user name is
used we can prevent issuing new tokens on an unsubscribe/resubscribe cycle.
If different user names are used but common domain names (an attack which
many people could mount) we could recognize that with somewhat more difficulty,
and mark those domains as special.  Most people would have trouble getting
lots of different accounts with different domain names.  Eric Hughes maxim,
"all crypto is economics", applies here.  We can easily make it much more
difficult for flooding attacks to occur.

Hal