5 ms·

Is there any way to tell if a particular email is spam, without knowing the content or the sender of said email?

by yen223 13y ago

olalonde 13y ago
Maybe homomorphic encryption[0] could help here? Here's what comes up after a quick Google: http://www.eweek.com/c/a/Security/IBM-Uncovers-Encryption-Scheme-That-Could-Improve-Cloud-Security-Spam-Filtering-135413/ http://www.eweek.com/c/a/Security/IBM-Uncovers-Encryption-Sc... [0] http://en.wikipedia.org/wiki/Homomorphic_encryption http://en.wikipedia.org/wiki/Homomorphic_encryption
ChuckMcM 13y ago
If all your actual email is encrypted then by definition spam is the unencrypted stuff. A long time ago in a different galaxy I built a PGP MTA (based on sendmail at the time) which only forwarded mail that was encrypted, and as expected it was spam free, all though these days spammers just might go to the trouble of sending it encrypted if they thought it would get through.
- csense 13y ago
  I'm guessing that people who only accept encrypted mail are such a small minority, you're probably safe from most spammers if you do this. People who target your company and employees specifically, maybe not. But even this could be made more difficult for spammers by only allowing access to your public key directory from trusted IP ranges. It's also a great idea from the standpoint of giving an encryption policy teeth. I'm thinking of a company where the official line is "We encrypt all our email," but then some IT screwup results in having half the company sending each other cleartext email for several quarters before anybody notices. If your mailservers reject unencrypted mail, the above scenario can't happen, because presumably people will notice when everyone's mail starts getting discarded, and it'll be fixed very quickly.
- rdl 13y ago
  Even just setting START TLS REQUIRED might solve your spam problem, as long as only a tiny minority of people did it. That would have the added benefit of protecting you from Yahoo Mail users, the FBI, and such. At this point, I'd consider NOT using START TLS for your MTA to be nearly as irresponsible as not using ssh instead of telnet/rsh, or not using secure passwords. It correctly pushes all the pain onto the sysadmin (and a very tiny amount of pain), rather than end users.
  - bigiain 13y ago
    Do you know if a successful response to a START TLS command endured end-to-end TLS secured mail transport? I kinda doubt it - if for some reason your outgoing mail server connects to one of my secondary/relaying MX servers, I don't think there's any way for you to ensure that server bothers trying to set up a TLS session when it relays my mail(which I guess is mostly my problem/fault) - and similarly, if your ISP requires you to send mail via their SMTP servers (blocking port 25 isn't uncommon here) - I don't think you've got any say in whether or not that server requires TLS? (I know - I really should go and look this up myself…)
    - rdl 13y ago
      Usually people do not block 465 or 587 (if they do, they really really suck, and you need to VPN through that network anyway). For outgoing mail, you just do STARTTLS directly to your own smarthost over those ports.
clicks 13y ago
It could be a crowd-sourced effort -- e-mails that are spam you mark as spam and upload to some repo that is maintained by someone (a la Adblock). You keep your filters updated, and run your e-mail against the filters file.
- ricardobeat 13y ago
  That's how current spam filters work, every mail server reports back to some central directories which keep tab on IPs, common subjects, keywords and other characteristics used to calculate spam scores. The trick is doing that without giving out access to message content to third-parties. http://en.wikipedia.org/wiki/The_Spamhaus_Project http://en.wikipedia.org/wiki/The_Spamhaus_Project
  - rdl 13y ago
    You can free ride -- receive the spam directory but not contribute. As long as your mail is representative of the overall mail corpus, that works out fine.
- eru 13y ago
  That would be trivially defeated by sending everyone a slightly different spam email. (And if your encryption doesn't produce totally different files for slight changes in plaintext, it doesn't deserve that name.)
  - ricardobeat 13y ago
    I think you're replying to the wrong comment tree. Spammers already do this today, by including your name and other data in the message, and varying the wording, but that is not enough to fool the spam blockers, and there are diminishing rewards as your addresses/IPs begin to get marked as source of spam regardless of the content.
    - eru 13y ago
      No, it was the right one. I was thinking about the grand-grand-fathers suggestion in relation to encryption.
rdl 13y ago
No need to go that far. The provider just pushes code to a server the user controls. The code is trusted to just do mail server with no backdoor for the service provider to get data. That code executes in an environment trusted by the data owner, so the code is allowed to see the data and process based on it. I'm kind of tempted to do this, since we largely do this kind of thing for VPN already. Doing it with trusted execution lets the code execute on hardware owned by the service provider, but that is technically difficult and hard to prove. Just letting you run code on your own virtual machine host is a lot easier (could be a third party virtualization provider if that is what you want, but I'd at least go for colo of my own server, if not on premise.)
- clicks 13y ago
  > I'm kind of tempted to do this, since we largely do this kind off thing for VPN already. Take advantage of the zeitgeist rdl. You can look forward to at least one customer if you decide to go ahead with this. (me). :)
betterunix 13y ago
Yes, but it is still very much in the research stages, not quite ready for real-world deployment: https://en.wikipedia.org/wiki/Fully_homomorphic_encryption https://en.wikipedia.org/wiki/Fully_homomorphic_encryption https://en.wikipedia.org/wiki/Secure_multi-party_computation https://en.wikipedia.org/wiki/Secure_multi-party_computation There is a huge amount of interest among cryptography researchers and from DARPA and the NSF.

Is there any way to tell if a particular email is spam, without knowing the content or the sender of said email?

by yen223 13y ago

Is there any way to tell if a particular email is spam, without knowing the content or the sender of said email?

Maybe homomorphic encryption[0] could help here? Here's what comes up after a quick Google: http://www.eweek.com/c/a/Security/IBM-Uncovers-Encryption-Scheme-That-Could-Improve-Cloud-Security-Spam-Filtering-135413/ http://www.eweek.com/c/a/Security/IBM-Uncovers-Encryption-Sc... [0] http://en.wikipedia.org/wiki/Homomorphic_encryption http://en.wikipedia.org/wiki/Homomorphic_encryption

ChuckMcM 13y ago

If all your actual email is encrypted then by definition spam is the unencrypted stuff. A long time ago in a different galaxy I built a PGP MTA (based on sendmail at the time) which only forwarded mail that was encrypted, and as expected it was spam free, all though these days spammers just might go to the trouble of sending it encrypted if they thought it would get through.

csense 13y ago
I'm guessing that people who only accept encrypted mail are such a small minority, you're probably safe from most spammers if you do this. People who target your company and employees specifically, maybe not. But even this could be made more difficult for spammers by only allowing access to your public key directory from trusted IP ranges. It's also a great idea from the standpoint of giving an encryption policy teeth. I'm thinking of a company where the official line is "We encrypt all our email," but then some IT screwup results in having half the company sending each other cleartext email for several quarters before anybody notices. If your mailservers reject unencrypted mail, the above scenario can't happen, because presumably people will notice when everyone's mail starts getting discarded, and it'll be fixed very quickly.
rdl 13y ago
Even just setting START TLS REQUIRED might solve your spam problem, as long as only a tiny minority of people did it. That would have the added benefit of protecting you from Yahoo Mail users, the FBI, and such. At this point, I'd consider NOT using START TLS for your MTA to be nearly as irresponsible as not using ssh instead of telnet/rsh, or not using secure passwords. It correctly pushes all the pain onto the sysadmin (and a very tiny amount of pain), rather than end users.
- bigiain 13y ago
  Do you know if a successful response to a START TLS command endured end-to-end TLS secured mail transport? I kinda doubt it - if for some reason your outgoing mail server connects to one of my secondary/relaying MX servers, I don't think there's any way for you to ensure that server bothers trying to set up a TLS session when it relays my mail(which I guess is mostly my problem/fault) - and similarly, if your ISP requires you to send mail via their SMTP servers (blocking port 25 isn't uncommon here) - I don't think you've got any say in whether or not that server requires TLS? (I know - I really should go and look this up myself…)
  - rdl 13y ago
    Usually people do not block 465 or 587 (if they do, they really really suck, and you need to VPN through that network anyway). For outgoing mail, you just do STARTTLS directly to your own smarthost over those ports.

clicks 13y ago

It could be a crowd-sourced effort -- e-mails that are spam you mark as spam and upload to some repo that is maintained by someone (a la Adblock). You keep your filters updated, and run your e-mail against the filters file.

ricardobeat 13y ago
That's how current spam filters work, every mail server reports back to some central directories which keep tab on IPs, common subjects, keywords and other characteristics used to calculate spam scores. The trick is doing that without giving out access to message content to third-parties. http://en.wikipedia.org/wiki/The_Spamhaus_Project http://en.wikipedia.org/wiki/The_Spamhaus_Project
- rdl 13y ago
  You can free ride -- receive the spam directory but not contribute. As long as your mail is representative of the overall mail corpus, that works out fine.
eru 13y ago
That would be trivially defeated by sending everyone a slightly different spam email. (And if your encryption doesn't produce totally different files for slight changes in plaintext, it doesn't deserve that name.)
- ricardobeat 13y ago
  I think you're replying to the wrong comment tree. Spammers already do this today, by including your name and other data in the message, and varying the wording, but that is not enough to fool the spam blockers, and there are diminishing rewards as your addresses/IPs begin to get marked as source of spam regardless of the content.
  - eru 13y ago
    No, it was the right one. I was thinking about the grand-grand-fathers suggestion in relation to encryption.

rdl 13y ago

No need to go that far. The provider just pushes code to a server the user controls. The code is trusted to just do mail server with no backdoor for the service provider to get data. That code executes in an environment trusted by the data owner, so the code is allowed to see the data and process based on it. I'm kind of tempted to do this, since we largely do this kind of thing for VPN already. Doing it with trusted execution lets the code execute on hardware owned by the service provider, but that is technically difficult and hard to prove. Just letting you run code on your own virtual machine host is a lot easier (could be a third party virtualization provider if that is what you want, but I'd at least go for colo of my own server, if not on premise.)

clicks 13y ago
> I'm kind of tempted to do this, since we largely do this kind off thing for VPN already. Take advantage of the zeitgeist rdl. You can look forward to at least one customer if you decide to go ahead with this. (me). :)

betterunix 13y ago

Yes, but it is still very much in the research stages, not quite ready for real-world deployment: https://en.wikipedia.org/wiki/Fully_homomorphic_encryption https://en.wikipedia.org/wiki/Fully_homomorphic_encryption https://en.wikipedia.org/wiki/Secure_multi-party_computation https://en.wikipedia.org/wiki/Secure_multi-party_computation There is a huge amount of interest among cryptography researchers and from DARPA and the NSF.