This last example demonstrates the dramatic difference between URL-based filtering and
content-based filtering. The structure of the Internet makes it very easy to block all sites ending in .il, but extremely hard, if not impossible, to block all sites containing content about
Israel, a project that our data indicates was never seriously undertaken within the country. It
may be that the purpose of blocking ‘‘.il’’ was more a statement about the Syrian and UAE
view on Israel, rather than an attempt to prevent its citizens from discovering particular information. Also, the block likely operates in both directions: someone from a ‘‘.il’’ address may
have a hard time accessing content in the UAE as a result of the filtering there.
The Role of Commercial Software in State-Mandated Internet Filtering
Commercial services, including the U.S.-based Secure Computing’s SmartFilter, Websense,
and Fortinet, appear to assist, or to have assisted, states that filter with the implementation
and management of block lists. These services provide extensive lists of URLs categorized
using proprietary methods. The commercial services typically fall in the middle of a spectrum
between manual and automated filtering. The URL for a site found to contain content related
to gambling will be offered as a digital update to the ‘‘gambling’’ block list of those states subscribing to the filtering services’ lists.
For topics such as pornography or drugs, few states appear to invest the resources
required to maintain active block lists where they can procure a list from commercial Internet
filtering companies. The challenge in doing so is compounded by multiple means of typically
accessing such Web sites—http://www.norml.com, http://norml.com, http://www.norml.org,
and http://209.70.46.40 all bring the user to the home page of the National Organization
for the Reform of Marijuana Laws. Each of these means of accessing the site (and others)
must be added to the block list in order to block citizens’ access in a thorough manner.
The task is further complicated when site operators realize that they are being filtered and
attempt to evade simple filtering techniques by changing their URLs; for instance, Iran
has blocked the site at www.pglo.org, but in a subsequent test, it had not blocked the
same content on www.pglo1.org.
18 Additionally, since filtering on a national scale requires
complex infrastructure, making sure that the same list of blocked sites is present on each
machine performing the filtering, at either a centralized or ISP-specific level, is no simple
task.
A state subscribing to such a service is limited to the categories made available by the
commercial software providers. While generally useful for content targeted according to the
common desires of parents, schools, and companies in the West (such as ‘‘pornography,’’
‘‘drugs,’’ or ‘‘dating’’), these products also include broad categories such as ‘‘religion’’ and
‘‘politics’’ that are not fine-tuned enough to match state-specific goals. These categories will
not, in the off-the-shelf version of the software, include filtering of content critical of Islam or
opposing the government of Vietnam. To account for the generality of these categories, each
38
Jonathan Zittrain and John Palfrey
content-based filtering. The structure of the Internet makes it very easy to block all sites ending in .il, but extremely hard, if not impossible, to block all sites containing content about
Israel, a project that our data indicates was never seriously undertaken within the country. It
may be that the purpose of blocking ‘‘.il’’ was more a statement about the Syrian and UAE
view on Israel, rather than an attempt to prevent its citizens from discovering particular information. Also, the block likely operates in both directions: someone from a ‘‘.il’’ address may
have a hard time accessing content in the UAE as a result of the filtering there.
The Role of Commercial Software in State-Mandated Internet Filtering
Commercial services, including the U.S.-based Secure Computing’s SmartFilter, Websense,
and Fortinet, appear to assist, or to have assisted, states that filter with the implementation
and management of block lists. These services provide extensive lists of URLs categorized
using proprietary methods. The commercial services typically fall in the middle of a spectrum
between manual and automated filtering. The URL for a site found to contain content related
to gambling will be offered as a digital update to the ‘‘gambling’’ block list of those states subscribing to the filtering services’ lists.
For topics such as pornography or drugs, few states appear to invest the resources
required to maintain active block lists where they can procure a list from commercial Internet
filtering companies. The challenge in doing so is compounded by multiple means of typically
accessing such Web sites—http://www.norml.com, http://norml.com, http://www.norml.org,
and http://209.70.46.40 all bring the user to the home page of the National Organization
for the Reform of Marijuana Laws. Each of these means of accessing the site (and others)
must be added to the block list in order to block citizens’ access in a thorough manner.
The task is further complicated when site operators realize that they are being filtered and
attempt to evade simple filtering techniques by changing their URLs; for instance, Iran
has blocked the site at www.pglo.org, but in a subsequent test, it had not blocked the
same content on www.pglo1.org.
18 Additionally, since filtering on a national scale requires
complex infrastructure, making sure that the same list of blocked sites is present on each
machine performing the filtering, at either a centralized or ISP-specific level, is no simple
task.
A state subscribing to such a service is limited to the categories made available by the
commercial software providers. While generally useful for content targeted according to the
common desires of parents, schools, and companies in the West (such as ‘‘pornography,’’
‘‘drugs,’’ or ‘‘dating’’), these products also include broad categories such as ‘‘religion’’ and
‘‘politics’’ that are not fine-tuned enough to match state-specific goals. These categories will
not, in the off-the-shelf version of the software, include filtering of content critical of Islam or
opposing the government of Vietnam. To account for the generality of these categories, each
38
Jonathan Zittrain and John Palfrey