of the installations of filtering software we observed appears to allow a state to augment a
commercial block list with its own URLs.
Aside from such fine tuning, however, states using commercial filtering services must
choose between allowing or blocking all URLs within a category. For example, a previous
version of the SmartFilter service provides the choice of blocking or allowing all URLs in the
‘‘anonymizer/translator’’ category. Even though a state may wish to block anonymizers in
order to prevent circumvention, that same state may wish to preserve access to translators
as a useful tool.
19 Language presents an additional problem, as all the commercial filtering
software we observed is produced by American companies. The blocking in states using
these commercial filters therefore tilts heavily toward evaluating—and in turn prompting
blocks of—English-language sites. This tilt leans precisely the opposite way from the tilt of
those states that develop their own blocking systems, which generally seek to block content
in local languages more than content in foreign languages. In some of the states using commercial filtering software, we have observed heavy filtering of English language content
in some categories, while the same content appeared to be freely available in the local
language—likely the inverse of what the state was seeking to accomplish through its filtering
regime.
20
When commercial filtering software is in use, a given second-level domain—for instance,
cnn.com—may include some sites that are blocked and other sites that are unblocked. Our
testing of SmartFilter has determined that the software attempts a more exact match first,
and in its absence falls back to categories assigned more generally to areas of a domain or
the domain itself. For instance, SmartFilter categorizes the Sports Illustrated home page at
sportsillustrated.cnn.com as ‘‘sports.’’ The default categorization for any Web page located
within this site, as shown by the category SmartFilter assigns to a request for http://
sportsillustrated.cnn.com/does_not_exist, will also be ‘‘sports.’’ However, the page for the
most recent swimsuit edition, at http://sportsillustrated.cnn.com/features/2006_swimsuit, is
categorized as ‘‘provocative attire/mature.’’ Thus, it appears that any Web page within the
Sports Illustrated site will, logically enough, be assigned to the ‘‘sports’’ category whether or
not SmartFilter has analyzed the content of the page, unless this default has been overridden
with a page-specific categorization.
21
Commercial filtering software may alleviate some of the difficulties of filtering presented by
the technical structure of the Internet. Our data show that states using such software are
much less likely to miss alternative means of accessing blocked sites, for instance, visiting
http://ifex.org to get around a block of http://www.ifex.org, as was possible in Vietnam during
our testing. Commercial software companies have refined their filtering techniques to anticipate, detect, and prevent these relatively simple methods of evading blocking. There are
others, as Deibert and Rohozinski note in chapter 6 of this volume, who seek to achieve just
the opposite. A game of cat and mouse is well underway.
Internet Filtering: The Politics and Mechanisms of Control
39
commercial block list with its own URLs.
Aside from such fine tuning, however, states using commercial filtering services must
choose between allowing or blocking all URLs within a category. For example, a previous
version of the SmartFilter service provides the choice of blocking or allowing all URLs in the
‘‘anonymizer/translator’’ category. Even though a state may wish to block anonymizers in
order to prevent circumvention, that same state may wish to preserve access to translators
as a useful tool.
19 Language presents an additional problem, as all the commercial filtering
software we observed is produced by American companies. The blocking in states using
these commercial filters therefore tilts heavily toward evaluating—and in turn prompting
blocks of—English-language sites. This tilt leans precisely the opposite way from the tilt of
those states that develop their own blocking systems, which generally seek to block content
in local languages more than content in foreign languages. In some of the states using commercial filtering software, we have observed heavy filtering of English language content
in some categories, while the same content appeared to be freely available in the local
language—likely the inverse of what the state was seeking to accomplish through its filtering
regime.
20
When commercial filtering software is in use, a given second-level domain—for instance,
cnn.com—may include some sites that are blocked and other sites that are unblocked. Our
testing of SmartFilter has determined that the software attempts a more exact match first,
and in its absence falls back to categories assigned more generally to areas of a domain or
the domain itself. For instance, SmartFilter categorizes the Sports Illustrated home page at
sportsillustrated.cnn.com as ‘‘sports.’’ The default categorization for any Web page located
within this site, as shown by the category SmartFilter assigns to a request for http://
sportsillustrated.cnn.com/does_not_exist, will also be ‘‘sports.’’ However, the page for the
most recent swimsuit edition, at http://sportsillustrated.cnn.com/features/2006_swimsuit, is
categorized as ‘‘provocative attire/mature.’’ Thus, it appears that any Web page within the
Sports Illustrated site will, logically enough, be assigned to the ‘‘sports’’ category whether or
not SmartFilter has analyzed the content of the page, unless this default has been overridden
with a page-specific categorization.
21
Commercial filtering software may alleviate some of the difficulties of filtering presented by
the technical structure of the Internet. Our data show that states using such software are
much less likely to miss alternative means of accessing blocked sites, for instance, visiting
http://ifex.org to get around a block of http://www.ifex.org, as was possible in Vietnam during
our testing. Commercial software companies have refined their filtering techniques to anticipate, detect, and prevent these relatively simple methods of evading blocking. There are
others, as Deibert and Rohozinski note in chapter 6 of this volume, who seek to achieve just
the opposite. A game of cat and mouse is well underway.
Internet Filtering: The Politics and Mechanisms of Control
39