[liberationtech] an Audit on Bing's China Censorship (or, an Independent Transparency Report)

[夏楚]

Dear all (cc greatfire.org, Rebecca MacKinnon and Jason Q. Ng),

I just published a thorough research on Bing's censorship for China. See
the paper at http://goo.gl/0XmxpF<https://www.google.com/url?q=http://goo.gl/0XmxpF&usd=2&usg=ALhdy29bKXUxqk6998VcqQRISp5r2kZNhw>and
the spreadsheet at
http://goo.gl/0XmxpF. The Twitter link is at
here<https://twitter.com/SummerAgony/status/439318305996689408>
.

Below is the Abstract

Microsoft has long been known of practicing censorship in China with its
Bing search engine, but little is known on how Bing's China-related
filtering actually works. In this thorough study, we examined Bing's SERP
(search engine result page) for a large body of (30,000+) sensitive and
non-sensitive query terms, queried from inside China and outside China.
Comparing and examining these results, plus querying with special search
operators, surfaced unprecedented details on Bing's China filtering
practice. In a sense, this is an independent "Transparency Report".

This study is motivated by the scandal in which Microsoft was accused of
practicing China censorship globally, first broken by GreatFire.org in
February, but denied by
Microsoft.

Main findings from this research:

   - Bing has a list of ``forbidden'' terms that no results are shown. 149
   such terms are identified in this study.
   - Bing has a blacklist of websites that it never shows to China users.
   329 such websites are identified in this study.
   - Bing has a huge blacklist of URLs that it never shows to China users.
   1593 URLs are identified in this study, which is only the tip of the
   iceberg.
   - ``Censorship leakage'', we observed censorship notices for 1710 unique
   query terms on the Chinese version of Bing from outside China.

In particular, I'd like to point out two issues.

   - *Bing's filtering is very broad*, for example, Bing filters the entire
   site for five major editions of Wikipedia: Japanese, French, German, Dutch
   and Swedish.
   - *Bing has extensive filtering rules for domestic sites *like
   blog.163.com, baidu.com, blog.sina.com.cn, 360doc.com, ifeng.com,
   douban.com etc. The content filtered by Bing still exists on these
   domestic sites and are not even filtered by domestic search engines like
   Baidu.

Overall, from these findings, my conclusion is that Bing has not lived up
to their promise of "support for freedom of access to information". For
example, removing the entire sites of ja.wikipedia.org, fr.wikipedia.org,
de.wikipedia.org, nl.wikipedia.org and sv.wikipedia.org does not support
Bing's claim of "if we are required to implement the request, we will do so
narrowly". Also, the fact that Bing filters vast amount of domestic content
and *more than Baidu does* is utterly saddening and inexcusable.

Link to this report, actual lists identified, and future updates can all
be found at the master spreadsheet http://goo.gl/qzAaVe.

My requests to Microsoft:

   - Explain why Bing censors so much content, including domestic content
   which are not even censored by Chinese companies.
   - Audit the actual procedure and handling of censorship in China, to
   verify that all government requests are scrutinized by proper personel. In
   particular, investigate the approval process for the domestic content.
   - Present the proof that removing entire sites off Bing in China is
   based on applicable law. In particular, the removal of the entire Japanese,
   French, German, Dutch and Swedish Wikipedia sites.
   - Do not attempt to hide censorship details, e.g. by altering censorship
   notices, altering how search operators like \verb|site:| and \verb|url:|
   work, or tweaking ranking algorithms.


Sharing and comments are very welcome. Don't hesitate to email me if you
have further questions.

--
Xia Chu (Twitter: @summer.agony)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.stanford.edu/pipermail/liberationtech/attachments/20140228/18c79191/attachment.html>