As of this writing, the websites of the New Braunfels Republican Women, the Kiddie Kollege Nursery School, the Freemasons’ District Grand Lodge of East Africa, the Weston Community Children’s Association, and the Rotary Club of Midland, Ontario are all categorized as “pornography” by Blue Coat Internet blocking software. Blue Coat Systems provides a lookup form that you can use to see how their system categorizes a website; you can see the screen captures for the categorizations of the above-listed sites here, here, here, here, and here. (The complete list of miscategorized sites that we found, is listed at the end of this report. Unless otherwise specified, all sites mentioned in this report, are currently classified by Blue Coat as “pornography.”)

Blue Coat Systems is based in Sunnyvale, California and manufactures a range of Internet filtering products, often used in schools and workplaces. On any network where the software is installed and configured to block sites categorized as “pornography,” users will be denied access to such sites.

The company was previously the subject of an in-depth report published by the Citizen Lab in January 2013, describing the use of Blue Coat devices in Syria and Burma for mandatory country-wide Internet censorship. Several dozen other Blue Coat installations were found in 19 “countries of interest” that were known to have a history of Internet censorship and human rights violations, such as China, Saudi Arabia, and Venezuela. As discussed in the report, it is not currently against US law for American companies to sell censorship software to foreign governments, although the practice is controversial. However, the use of the filtering appliances on Syrian networks prompted concerns from the US Department of Commerce about whether Blue Coat had sold those devices in violation of U.S. sanctions against Syria. Blue Coat claimed that the products had been bought by an entity in Dubai and re-sold to ISPs in Syria without Blue Coat’s knowledge.

Citizen Lab’s report concluded by posing several questions to Blue Coat, such as: “What human rights policy commitments and due diligence measures does Blue Coat Systems have in place concerning the development and sales of its products and services?” The report stated: “We commit to publishing in full Blue Coat System’s reply.” Citizen Lab did not receive a response from the company.

Compared to the practice of selling “censorware” to foreign governments with poor human rights records, there has been far less controversy over the simple miscategorization of websites as “pornography” by Internet blocking companies. There have been no public reports to date on sites that have been miscategorized by Blue Coat specifically, although there have been reports about other products, such as Citizen Lab’s post in November 2013 about sites miscategorized by SmartFilter. (As discussed below, a few sites actually showed up on both lists—miscategorized as “pornography” by both Blue Coat and SmartFilter—for reasons that are not clear.)

Some of the entities whose websites were classified as “pornography,” such as the Kiddie Kollege Nursery School, or the Lotus Valley Public School of Haryana, India, would likely be scandalized if their customers or the public came to believe, as a result of misclassification by Blue Coat, that their websites really were hosting “pornography.” (In the case of other miscategorized websites such as http://www.dicksofamerica.org, the proprietors would most likely be bemused by error.)

As with the SmartFilter report, we tested a large number of URLs against Blue Coat’s filtering system using a combination of ad hoc spidering and URL lists. After identifying a subset of URLs classified by Blue Coat as “pornography,” we reviewed the site contents to identify classifications that appeared to be errors. Some were obvious errors, and others were “borderline cases,” where the site contained no true pornography but it is plausible that a reviewer might have mistakenly applied the label after finding the site’s contents to be “inappropriate.” Blue Coat’s site lookup form gives their working definition of “pornography:”

Sites that contain sexually explicit material for the purpose of arousing a sexual or prurient interest.

None of the sites we counted as “errors,” contained any material that we could find which fit that definition.

While Blue Coat’s site lookup form does not give a reason for the categorization of particular sites, there were some sites for which we could infer a possible reason for the miscategorization. Only one site in our list, http://www.againstchildporn.org, contained an obvious keyword (“porn”) in the domain name which would have likely caused the site to be misclassified. A few other sites such as http://www.gambiawomen.org (“women”) and http://www.dicksofamerica.org (“dicks”) contained keywords which may have triggered the “pornography” categorization. (The latter site does in fact reference the humorous use of the word “dick,” although it is also the site owner’s first name.) Several other sites in our list contained content on the front page that may have caused Blue Coat’s system to categorize the site as “pornography:”

Meanwhile, there are a few sites in our sample categorized by Blue Coat as “pornography”, whose content is not pornographic, but may have been considered sufficiently “questionable” that it’s plausible that a human reviewer may have miscategorized it:

It would be natural to assume that, in the cases of the sites listed above, the reason they were mistakenly blocked as “pornography” by Blue Coat was either because of the keywords that appear on the page or in the domain name, or because of their “borderline” content. Certainly out of the sites that we found to be blocked in error, a high proportion of them contain “borderline” content and probable-trigger keywords, higher than would be expected if we had selected sites from the web at random.

On the other hand, many other websites such as Singing Bowls of Tuscon Shambhala, The Sumter Community Concert Band, and Reunion Ministries, are also categorized by Blue Coat as “pornography” despite having no pornographic content and no apparent keywords or content that could have triggered even an accidental misclassification as “porn.”

So, it is possible that a certain number of sites are wrongly blocked by Blue Coat as a result of some pseudo-random process, and a certain number of those sites will just happen to have keywords or questionable content on them—but the keywords and the content may not be the cause of the blocking. (It is impossible to know for sure, since Blue Coat’s lookup form does not give reasons for a site’s classification.)

A small number of sites in our sample were miscategorized by both SmartFilter and Blue Coat as “pornography.” They are:

The fact that these sites are miscategorized by both products is something of a mystery. If either SmartFilter or Blue Coat were simply miscategorizing a random subset of non-pornographic websites as “pornography,” then we would expect the overlap to be much smaller. (Out of roughly 150 websites listed in each report, eight were also miscategorized by the other product, or about six percent. But despite all the errors made by both products, the proportion of non-pornographic websites that either product categorizes as “pornography” is generally far less than six percent. So the high overlap appears to be due to something other than chance.) However, there appears to be no common factors that would cause these websites to have a greater-than-average chance of being miscategorized as “pornography.” None of them contain any questionable content or keywords likely to trigger a “pornography” categorization (except possibly for the “New Braunfels Republican Women”).

The sampling of sites that we found to be incorrectly classified as “pornography” by Blue Coat, is below. Each site is listed together with a link (labeled “sc”) to a screen capture showing the site classified as “pornography” on Blue Coat’s lookup form. The classifications are current as of this writing; Blue Coat may change the categorization of these sites in the future.

In our testing, we also found some sites such as http://www.iawa-web.org, the International All-round Weightlifting Association, which appeared to have been hacked to insert sexually explicit text in a sidebar (although most of the site’s original content was also left intact). We did not count such sites as “incorrectly classified” when Blue Coat listed them as “pornography,” due to the presence of the sexually explicit text. However, we believe the best practice on Blue Coat’s part would have been to:

1) give the sites a special classification indicating that the site appeared to have been hacked temporarily;

2) notify the site owners where possible; and

3) re-check the site periodically to see if the problem had been fixed and the text had been removed. In cases where we could find contact information for the sites in question, we sent them a message indicating that they had been hacked.

Our testing indicated that these non-pornographic websites, despite being classified as “pornography,” were not generally blocked in countries that we had found to be using Blue Coat Internet censorship appliances. As far as we can tell, the United Arab Emirates, as mentioned in our previous report on SmartFilter, is using Blue Coat’s hardware and software, but filtering websites on SmartFilter’s list. (Therefore, the only sites in our list above that were blocked in the United Arab Emirates are the sites that are also categorized as “pornography” in SmartFilter’s database.)

However, as a follow-up to our post about the number of sites miscategorized by SmartFilter, our tests with Blue Coat show that this is not a problem limited to a single product. We should be skeptical of any company’s claims that they are able to categorize much of the web accurately, or that their rate of “collateral damage” is very low.

Acknowledgements

The author thanks Citizen Lab for their support of this project and for their feedback and assistance during the research process.