Distributed Spam Detection

A reader writes "There's an interesting project at SourceForge, called, "Vipul's Razor", that uses a gnutella like system to let users exchange spam "signatures" to filter spam. I work at an ISP in Ottawa, we have been using it for last two weeks to stop bulk of spam coming to our POP3 accounts. More impressively, it hasn't tagged any valid mail as spam yet. Here's the scoop from its webpage: "Vipul's Razor is a distributed, collaborative, spam detection and filtering network. Razor establishes a distributed and constantly updating catalogue of spam in propagation. This catalogue is used by clients to filter out known spam. On receiving a spam, a Razor Reporting Agent (run by an end-user or a troll box) calculates and submits a 20-character unique identification of the spam (a SHA Digest) to its closest Razor Catalogue Server. The Catalogue Server echos this signature to other trusted servers after storing it in its database. Prior to manual processing or transport-level reception, Razor Filtering Agents (end-users and MTAs) check their incoming mail against a Catalogue Server and filter out or deny transport in case of a signature match."" Cool idea. I'm up around 80% spam a day on my main mail account. Might be worth a try.

Comments

SpamBouncer

[joib]

I'm personally using SpamBouncer, a procmail-based spam filter. Works fine for me.

Great use of p2p

[astrashe]

This is a great use of p2p -- something that doesn't involve piracy. I wish I had heard of it before.

Are there any other innovative non-piracy p2p apps out there that we should know about?

Re:Great use of p2p

[Sarcasmooo!]

Just because most people on a P2P network use it for piracy, it doesn't become a pirate-app. I can, and have, used programs that are under attack by the RIAA do download speeches, text documents, etc. At the early point of the 2000 Nader campaign, when he couldn't get 30 seconds of time on M$NBC (much less a place in the debates later on), I used Napster and Scour to find speeches he's given. And when the Department of Commerce kicked of it's 'Safe Harbor' privacy program by failing to put the confidential information provided by the companies involved on a secure site, I downloaded the pages in a zip file despite the site being closed for a fix. Using programs like Scour, I found reading material on scientology, COINTELPRO, and more, all the way up until the day that lawsuits shut them down.

So...

[DagSverre]

...what stops this from being abused? Say I set up a box that automatically reports all mails on the most popular mailing lists as spam, effictively making the ISPs around the world start to filter out the mailing lists...

It's a great initiative, I really hope no troll out there takes my word on this and actually do this.

Authentication with servers?

[GlassUser]

I read some of the documentation, but I can't find details on a couple of questions. Do the servers authenticate with each other? It was implied, but how deep is it? Are the SHA signatures signed to the originating server (or client/trollbox) too? I think this kind of model is great, but if you don't have some nifty authentication/accountability, it can be wide open for abuse. I'm sure anyone reading slashdot can imagine a vengeful spammer flooding the network with bogus or malicious hashes.

Fighting spam

[Brian Kendig]

I'll post my usual public service announcements here:

SpamCop is a great service for reporting spam; just paste the spam message into the web form, and it'll automatically figure out where the smap came from and send complaints off to the appropriate people.

The Spam Bouncer is a procmail-based personal spam screening tool. It's got some interesting features, but I haven't used it in a long while.

The way I avoid spam is to have my mail client screen out any email which contains any of these phrases:

to be removed
to be permanently removed
to get removed
to get off the list
to get off this list
to be taken off
to remove yourself
removal instructions
remove in subject line
"remove" in subject line
remove in the subject
"remove" in the subject
'remove' in the subject
S.1618
S. 1618

This list by itself catches about 80% of the spam I get.

How do you compute a signature?

[cperciva]

As far as I can tell from a quick glance at this, it looks like the entire message body is being used to compute the signature. This isn't going to work very well -- over half of the spam I receive is "personalized", and that fraction is growing every day.

This could work very well, but we need some way of computing signatures which will be invariant across different copies of personalized spam for this to be effective.

SpamAssassin uses Razor

[wideangle]

From http://spamassassin.taint.org/:

SpamAssassin is a mail filter to identify spam.

Using its rule base, it uses a wide range of heuristic tests on mail headers and body text to identify "spam", also known as unsolicited commercial email.

The spam-identification tactics used include:

• header analysis: spammers use a number of tricks to mask their identities, fool you into thinking they've sent a valid mail, or fool you into thinking you must have subscribed at some stage. SpamAssassin tries to spot these.
• text analysis: again, spam mails often have a characteristic style (to put it politely), and some characteristic disclaimers and CYA text. SpamAssassin can spot these, too.
• blacklists: SpamAssassin supports many useful existing blacklists, such as mail-abuse.org, ordb.org or others.
• Razor: Vipul's Razor is a collaborative spam-tracking database, which works by taking a signature of spam messages. Since spam typically operates by sending an identical message to hundreds of people, Razor short-circuits this by allowing the first person to receive a spam to add it to the database -- at which point everyone else will automatically block it.

Once identified, the mail can then be optionally tagged as spam for later filtering using the user's own mail user-agent application.

SpamAssassin requires very little configuration; you do not need to continually update it with details of your mail accounts, mailing list memberships, etc. It accomplishes filtering without this knowledge, as much as possible.

Call your ISP and ask if they use it.

This is just a temporary solution.

[mrsam]

Spam generators have been trying to hash-bust these kinds of filters for years now. A four year spam generator automatically appends random junk at the end of the Subject header or at the tail end of the message, in order to defeat the early hash-based spam filters.

This is probably a 'fuzzy' hash function that should ignore minute variations. However, it goes without saying that if this hash-based spam filter becomes widespread, then the spammers will simply figure out how to hash-bust their way past it.

To have any hope of working over the long term, this kind of an approach must include the ability to distribute not just the hashes themselves, but the hash function as well, so that the hash function itself can be adjusted, when needed.

One way around potential abuse.

[chris_7d0h]

To eliminate the situation where one person posts a lot of "incorrect" signatures, a ranking system could be applied.
The thought goes like this.
A person submits a signature of "identified" spam mail to a "supernode" for ex. and the submission gets a ranking of 1. Each additional submission (by other users) increases the score by a number.

This way, there are several classifications which could be used to filter incoming mail. For the mail providers, they could opt for only removing mail matching signatures with a very high score (thus very likely these will be actual spam) or they could filter anything reported.

The purpose of allowing the use of classifications is that it will take longer time to get higher scores, since more people have to report the specific spam mail. Some people whish to eliminate things the least bit suspected, but mileage may vary.

Do you see a resemblance with the ./ moderation?

Re:Great use of p2p -- Wont work.

[DLG]

>> This wont work. All that will happen is that the spammers will just modify their spam programs to slightly modify each message they send out.

It will however require them to send each specific message separately rather than sending large cc's or using some sort of relay. That alone is a big step since right now most spammers can get away with sending a single email message and relying on an open relay to retransmit to a larger group.

Furthermore I have doubts that for the time being this project will concern spammers. Infact I am pretty sure spammers are not really interested in wasting their own time trying to spam people who consider spam a violation. It is more convenient to ignore those people (which is why they don't bother to check if you want spam or not before they send it to you).

DLG

Virus Detection

[doorbot.com]

This seems like it would be a great method for virus detection on a non-Windows machine. For those of you who run *nix mail servers which eventually filters down to Windows clients, having a mail tagged as viral would be nice to have it be immediately denied at the server. So I'm assuming all it would take is a smart admin to tag the email as spam, and then it will propagate around to the other servers (less than 1k would transfer!).

List of server-based spam filter systems

[tgeller]

A canonical list of server-based spam filtering systems is on the SpamCon Foundation site, along with other sysadmin resources.

Foreign spam removal

[wideangle]

For the many /.ers who:

a. Use Outlook secretly
b. Receive loads of foreign spam
c. Don't know any foreign languages
d. Don't have any foreign friends
e. Don't have any friends

This Outlook rule is for you!

Apply this rule after the message arrives
with
Ô or ¾ or Ç or or É or ½ or Í or ò or Ë or ® or Ä or ã or Ï or Ö or Ô in the subject or body
delete it
and stop processing more rules.

This blocks 99% of foreign spam. Sue Mosher wrote about other effective methods for killing spam in Outlook. Finally, before you reply saying "You dummy, that filter works in any client!" -- You're right.

Re:Great use of p2p -- Wont work.

[kevinank]

Interesting work, but I notice that you are only examining trigrams, and you are using an even weight factor. To improve selection you probably at least need to use variable weights (a fuzzy logic neural network rather than binary logic) and train the network with more sample spam.

I've been working on a similar project but using additional factors that help identify spam such as violations of the mail RFC's, and other header indicators, in addition to NLP. I have a prototype that I'm using to score all of my inbox e-mail and am using that to tune the weight factors and add in new factors as I encounter them. It would be interesting to combine your approach with mine I think, since I hadn't thought of analyzing trigrams.

Anyway, if you are interested send me an e-mail and I'll give you my current perl code.

Answers to some questions raised on slashdot.

[vipul_ved_prakash]

Hi,

Some of you point out that Razor's use of SHA-1 signatures can be defeated by introducing randomness in the message. This is true; SHA-1 will eventually be phased out and replaced by a fuzzy hashing mechanism like nilsimsa in future. [http://lexx.shinn.net/cmeclax/nilsimsa.html] [http://www.geocrawler.com/archives/3/2539/2001/7/ 0/6173567/] The protocol is structured to aid change of hashing algorithms seamlessly, without breaking the existing system. Regarding the possibility of poisoning the database, we are working on a reputation system that will assign credit to honest reporters. Once we have a critical mass of users, it would be hard for dishonest reporters to even join the reporting network, much less be able to mount a DOS attack. Some of these issues have been discussed on the razor-users mailing list. The list archives are located at [http://www.geocrawler.com/archives/3/2539/2001/] best, vipul.