The information contained in this document is accurate to the best of our knowledge, as of Feb 21 2003. However this is a complex and rapidly changing area - there may be errors that we have missed. If you find any factual errors, or you have any comments, please email <spamcopfeedback AT fastmail DOT fm>.
This paper is by Jeremy Howard, a director of FastMail.FM.
SpamCop BL is a system that is widely used by email providers for blocking email which may be spam. However, providers using this service are blocking up to 10,000 legitimate emails sent to their own customers, for each spam they block. SpamCop BL is a system that is open to abuse, and can be very inaccurate. SpamCop BL advises on their home page that production sites should not use the service - email providers that ignore this advice are causing problems for their own customers, while wasting the time of mail abuse departments that would be better spent fighting spammers.
SpamCop started out life as a marvelous spam reporting and notification service. I have commented numerous times on how useful and effective this service is.
Using the SpamCop notification service, users of any email system can easily report any email they receive. SpamCop then analyses the email, works out the likely true sender of the email, looks in the body to find any advertised email addresses or web sites, and then emails the administrators of all the systems involved to let them know about the problem. The SpamCop user is told who the notification will be sent to, and has the opportunity to remove any incorrectly targeted reports from the list. This service ensures that the right people know about abuse of their systems, and can quickly do something about it.
SpamCop quickly became very popular, and has leveraged its popularity to make profits from a for-fee email service, and from a "blocking list" which solicits donations.
SpamCop BL is the "SpamCop Blocking List". It is a list of IP addresses, which are the numbers that uniquely define computers connected to the Internet. This is what the SpamCop BL page has to say
This blocking list is somewhat experimental and should not be used in a production environment where legitimate email must be delivered. It is growing more stable and is used by many large sites now. However, SpamCop is aggressive and often errs on the side of blocking mail - users should be warned and given information about how their mail is filtered.
Email providers that decide that they are not "a production environment where legitimate email must be delivered" can check against this list any time a mail server delivers them email. SpamCop BL then provides instructions on how to reject emails that come from listed servers. Email providers that follow these instructions then block all email from a listed host.
So, the big question is: "how does SpamCop BL decide what to list?". Read on...
Unfortunately, SpamCop does not publish a current description of their algorithm anywhere. Therefore the only real way to find out what is listed, is to do some analysis as listings occur. SpamCop representatives have from time to time provided public information about the algorithm, and also maintain a page describing the algorithm. Unfortunately, the page has thus far not provided complete and accurate information, and the information provided by representatives has been inconsistent and sometimes incorrect. Another complexity is that the algorithm frequently changes.
Therefore, email providers that use SpamCop BL are stopping your customers from receiving emails based on criteria that you almost certainly do not fully understand.
We have attempted to reverse engineer the algorithm and this represents our current best understanding: (as of Feb 21 2003)
A SpamCop report occurs when a SpamCop notification system user reports a message as spam, and SpamCop's analysis results in it being connected with a particular email server. At FastMail.FM we have seen the following result in SpamCop reports
We have seen all of these sources of report, in roughly similar proportions. At FastMail.FM we observe on average about 30% of reports are correct. By way of example, here's a "spam" recently on file at SpamCop:
Content-Disposition: inline Content-Transfer-Encoding: 8bit Content-Type: text/plain MIME-Version: 1.0 X-Mailer: MIME::Lite 1.2 (F2.71; T1.001; A1.51; B2.12; Q2.03) Date: Sun, 1 Dec 2002 22:54:55 UT From: "Email Administrator" <> Reply-To: "FastMail Administrator" <webmasterATfastmail.fm> To: x Subject: URGENT: FastMail usage warning Message-Id: <2002_________________308F@www.fastmail.fm> Return-Path: email@example.com
A user had gone over their quota, we had told them so, and they reported that as spam! You'll also note that SpamCop munges the user's address with the false report, but not the service provider's address (I changed the '@' to an 'AT' above), allowing the service provider's address to be harvested by spammers' "robot" software.
In theory users making incorrect reports are banned from SpamCop, but in practice we've seen users getting warnings rather than bans, and SpamCop has no way to stop them just signing up with a new name (including automated multiple signups, automated fraudulent abuse reports, etc).
The listing algorithm, combined with the SpamCop reporting system (both described above), decide which servers are listed in the SpamCop BL. An analysis of the effectiveness of this algorithm for blocking email (the most appropriate analysis to use, given its name as a "blocking list") follows, in the categories of: accuracy, collateral damage, timeliness, and completeness.
Based on our experience, around 30-40% of spam reports targeted at our site are accurate. 100% of them result in SpamCop listings. A single report, whether accurate or not, will result in our site being blocked for 6 hours. That is because the report gets a score of 4 during the six hours after it is received. This is then divided by the number of 'good points'. SpamCop's collection of 'good points' is extremely ineffective - our site often has only around 200, despite our users sending and receiving millions of messages every day. So, bad=4 divided by good=200 = 0.02, which is at the listing threshold. So the site is listed for 6 hours.
Based on this, it would appear that on many occasions moderately large sites will get listed despite not being a source of any spam. Empirically, we have observed this behaviour. In general, the largest sites will have fewer problems, since their 'good score' will be large enough to avoid a single report causing a problem.
From time to time, a spammer will sign up for a FastMail.FM account, and will send spam. FastMail.FM has automatic monitors that will lock the account as soon as 100 recipients (for free accounts) have been sent to, and appropriate authorities are then contacted to take action. One of these spam recipients is likely to then report this to SpamCop.
Our analysis of SpamCop's statistics and our internal systems has let us to estimate that around 2.5% of spam recipients end up reporting to SpamCop. We have also calculated that the SpamCop BL "good score" is around 0.01% of the number of messages actually sent from that server during that time.
Now, let's assume that an accurate report has been sent, and it is over 6 hours since the report was received (so it counts for 1, not 4, "bad points"). The threshold is 0.02. Now, we estimate that 97.5% of spams have not been reported, so the actual number of spams sent for each report is around 50. At the threshold of 0.02 there must be no more than 50 "good points", which (using the 0.01% statistic) represents 500,000 actual sent messages. Thus, the actual proportion of spam sent from a host at the threshold is around 50/500,000 or 0.01%.
That means that email providers using the SpamCop BL are blocking providers where for each one spam they are blocking, they are blocking 10,000 legitimate emails from reaching their own customers!
At FastMail.FM, a free email user is blocked as soon as the 100th email recipient is reached in an hour. So generally a spammer will be blocked within a few seconds (they try to send hundreds of thousands in a short time - anything less and it's not worth their while). A SpamCop user then logs into their email the next day, sees a spam, and reports it. SpamCop then lists FastMail.FM, despite the spammer being locked 24 hours earlier.
In this case the system fails to actually block spam sources in a manner that stops people receiving spam.
There is a wide range of potential spam sources that SpamCop does not successfully block . One example is mail providers with many IP addresses over a range of different net-blocks. An example is Hotmail. Currently (as of Feb 21 2003) a Hotmail server is in SpamCop's blocking list - SpamCop provides the following listing details
18.104.22.168 listed in bl.spamcop.net.
Rationale: Recent spam increases spam score from 3.00 to 4.00: spam report ratio (0.038) exceeds threshold (0.020)
However, each email from Hotmail comes from a different sending server. For instance, yesterday FastMail.FM received email from 2,241 different Hotmail servers. So, if someone is unlucky enough to have their email sent through 22.214.171.124 right now, it will be blocked by users of SpamCop BL. Whereas if the reason for that block was really a spammer, that spammer's next message will probably go through a different server, and will not be blocked!
FastMail.FM is often the target of vengeful spammers. That's because we waste a lot of their time and money, for instance when they spend time setting up FastMail.FM accounts and find them locked almost immediately by our automated systems. They do everything they can to cause problems as a result, although most of the time our systems automatically identify and block whatever they try. One thing we have not been able to do much about, however, is a recent trick they have used to get us listed by SpamCop.
The spammers, we have discovered, send spam to under 100 people (thus defeating our locking system), but they send every spam to known anti-spam campaigners and SpamCop users! They know that almost all of the recipients will report to SpamCop, and get us listed. We know this has happened because when we have checked the send log on our server, we find the user in question has sent only one email, and that all the recipients are people that we know to be active in the anti-spam community.
Another example of abuse is reported here by a user at http://www.emaildiscussions.com :
I am listed on the abuse and technical whois records for our domain (A small government). Every day I have to fight with the idiots over at spamcop to keep us off their various forms of mailblocking strategies. We are a legit, spam-free government network, with extremely strict anti-spam policies. Hell, we have extremely strict private-use policies. Recently, a few people were sacked for abusing the mail system.
However, SpamCop insists on running a broken system that is totally open for abuse. Every loser with a grudge against the Government (i.e. everybody that pays tax) reports a mail to spamcop as being spam and as coming from us. The reporting mechanism is smooth, fast, free and has no consequences for the reporting user. The resulting work for us on the other hand is difficult, time consuming and very, very costly. We must take every report on spam or abuse seriously. So this involves tracking down the reported user, informing the manager, setting up a spot-audit team, going over to his/her PC, checking out the contents of the HDD, writing up reports, and the lot. This keeps about 4 people busy for a full working day.
We also have to work fast - i.e. drop what you are doing and respond to the issue. We have tried working with SpamCop on this issue, but to no avail. As far as we are concerned, in this case, the medicine is worse then the disease, and SpamCop is just another word for DDOS.
SpamCop should be shut down.
SpamCop's policies regularly differ in practice to what they document. For example, when Politech was listed by SpamCop BL, Julian of SpamCop said "Since complaints from their spam-victims don't seem to have any effect, perhaps complaints from their paying users will! If rackspace does not take action to stop this source of spam, it is quite possible that other, innocent rackspace customers will be affected again." Although SpamCop regularly state that the primary reason for blocking other IPs is not collateral damage, their comments and behaviour has repeatedly shown that they do believe in this system.
Another example of FAQ/reality mismatch... The FAQ says spam is "
". When a FastMail.FM user sent information to 30 people with publicly listed addresses with information that they thought was useful to those people (so this message was somewhat solicited, and definitely not automated), we were listed. When I reported this, the SpamCop representative said that the message was regarded as spam because it was sent to more than one person, one of whom felt (by reporting) that it wasn't solicited. Whilst I can certainly empathise with a definition of Unsolicited Commericial Email which covers that type of situation, this is not the stated definition on the Spamcop BL site. Email providers that use the Spamcop BL should be aware that the basis of listings in the blocking list is not the same as the criteria made publicly available.
Other examples of SpamCop problems:
On the other hand, SpamCop does not list itself even when they should. For instance, a lookup of Spamcop BL of one of Spamcop's IPs shows the following result
126.96.36.199 not listed in bl.spamcop.net.
188.8.131.52 is not listed, but should be.
Rationale: Recent spam increases spam score from 25.00 to 54.00: spam report ratio (0.116) exceeds threshold (0.020)
With SpamCop blocking competitors, and providing inconsistent and inaccurate information, it makes it a very uneven playing field for others in the anti-spam business. We hope that this paper helps to clarify some of the issues around the Spamcop Blocking List.