US Military Looks For Massive Spam Solution

Several users have pointed out a recent request to technology companies from the Defense Information System Agency for ideas on how to build an e-mail defense system to catch spam. The solution would have to scan about 50 million inbound messages a day across some 700 unclassified network domains. "Defense currently scans e-mails for viruses and spam coming into systems serving the military services, commands or units. DISA wants to extend the protection to the interface between the Internet and its unclassified network, the Non-classified Internet Protocol Router Network. The agency also wants the ability to scan all outbound e-mails from the 5 million users. [...] DISA's request ties in with recommendations that the Defense Science Board issued in April that said Defense is more vulnerable to cyberattacks because of its decentralized networks and systems. The board envisioned a major role for DISA in developing the architecture for enterprise-wide systems."

Comments

We need a whitelist that doesn't suck

[steveha]

The only solution is to make a system that uses a whitelist. But whitelists suck. So we need a whitelist that doesn't suck.

The first step is to have all the email clients start digitally signing emails. It is trivially easy to forge the headers on an email, so it would be stupid to trust them for identity information.

The second step is to have email servers check the identity against the whitelist. If the digital signature is invalid, or the credentials are forged (message was digitally signed, but the announced public key of the sender doesn't match) the message is trashed, with no error message sent. If the signature checks out, but the sender was not on the whitelist, the message bounces back to the sender, with an explanation ("you weren't on the whitelist, sorry").

Okay, but whitelists suck. If my best friend from college wants to track me down and send me an email, I want him to be able to do that; but I don't know his email so he's not on my whitelist. So, we need a solution to this problem.

My proposed solution is that your email server should advertise a list of ways that you will accept to bypass your whitelist for a message. One possible way: attach a micropayment of five cents. Another way: attach a certificate showing that your computer worked for an hour on some worthy problem like protein folding at home or something. Another way: here's a URL of a web page; it contains some riddle... attach the answer to your email. I'm sure you can think of other schemes to make it possible for a friend to bypass your whitelist while not enabling zombie Windows clusters to spray spam into your inbox.

There are other refinements possible. Your whitelist can accept, not just individual signatures, but "badges" from some organization. So, anyone from Mozilla.org can attach a Mozilla.org badge to their emails, and I can allow all Mozilla.org emails through. IEEE member badge, SourceForge.net badge, Apple.com badge, go nuts. Even an organization of "I Swear I Will Never Send Out Spam". The key with the badges is that, if you get kicked out of an organization, you have to lose access to the badge. One simple way would be for the check to be live: if you attach a Mozilla.org badge, the Mozilla.org server had better agree that your identity is one known to it.

The current email system is a "Default Permit" system (the #1 dumbest idea on this list). It has to change.

This system would run on the infrastructure we already have, with a few additions. You could have one account with the whitelist, and another account without... but the one with the whitelist is the only one that pages you, or whatever. The important thing is that this doesn't require everyone in the whole world to adopt it before it starts to become useful. Mailing lists would still work, because when you sign up for a mailing list you would add that mailing list identity to your whitelist (probably a badge, such that members of the mailing list are then cleared to email you directly, through the badge).

Someone may claim that validating public key signatures is computationally expensive. No, not compared to running complicated heuristics over the content of a message, trying to guess whether it's spam or not (SpamAssassin and other systems). With this system, the server doesn't attempt to classify a message. Either it passes the whitelist, it's bounced back to the sender, or it's deleted. Done.

Now, if you have found a hole in this idea, you will score bonus points by explaining how to fix it, not merely pointing out that I am an idiot.

steveha