New censorship on weibo

Back in October, we mentioned that weibo rolled out new semi-censorship. Apparently this is not enough. Now weibo has four different ways to censor keywords.

A. Explicit complete censorship

根据相关法律法规和政策,“[the blocked keyword]”搜索结果未予显示。

This translates into:

According to relevant laws, regulations and policies, search results for [the blocked keyword] can not be displayed.

Keyword example: 六四

Note: This is the good old way of censorship since the beginning of weibo.

B. Implicit complete censorship

抱歉,未找到“[the blocked keyword]”相关结果。

This translates into:

Sorry, no relevant results could be found for [the blocked keyword].

Keyword example: 江泽民

Note: This is a brand new form of censorship. Sina weibo used to admit what they censored. Now this message will also be shown to b keywords which actually have no results such as "dsfhadslfhadsljk".

C. Explicit semi-censorship


This translates into:

According to relevant laws, regulations and policies, some of the search results can not be displayed.

Note: This method is employed around October as mentioned in the blog post.

D. Implicit semi-censorship

When you search for the keywords, only some selected results are returned. With a message at the button, stating


This translates into:

In order to provide a diversity of results, we omitted some similar results. You could click here to view all results.

However, this message is shown to every keyword(except complete censored ones and those actually without any results). So you cannot determine the keyword is semi-censored from this message alone. When you click to view all results, sina weibo will append "&nodup=1" to the end of the search url. Now it will show the exact same message as b. Implicit complete censorship.

抱歉,未找到“[the blocked keyword]”相关结果。

This translates into:

Sorry, no relevant results could be found for [the blocked keyword].

Keyword example: 习近平

Note: This is a brand new way of censorship. It's really interesting to see how keywords are censored. For example, the censorship status of "习近平" went from a->not censored(?)->c->d->c. As I started to write this post, it was d, as I used web-archive to record it. When I'm writing this sentence 28 minutes later, however, it switched back to c. Another keyword "江泽民" went from a->b->a->b and stays at b. Our system could now detect censorship a, b(false positive of "dsfhadslfhadsljk" is inevitable), c. We might update our system to detect d if more keywords are censored this way in a prolonged time. Last but not least, this is merely censorship of searching on weibo. Much emphasis is also put on preventing users from posting sensitive weibo. Some keywords couldn't be posted on weibo in the first place, others will trigger a manual review by censors. Unfortunately, we haven't monitored that kind of censorship yet. But you could visit to view recently censored posts on weibo.

Update: Now Sina has the 5th way of censoring keywords-7 day delay, please read Sina Weibo Enacts New "7 Day Delay" Function for Sensitive Terms Following 18th Party Congress for detail.


More Blog Posts

Subscribe to our mailing list
Show content from Blog | Google+ | Twitter | All. Subscribe to our blog using RSS.

Thu, Sep 24, 2015

Apple blocked CNNIC CA months after MITM attacks

In March of this year, Google found unauthorized digital certificates for several Google domains. The root certificate authority for these domains was the China Internet Network Information Center (CNNIC). CNNIC was controlled by the Chinese government through the Ministry of Industry and Information Technology and is now under the management of the Cyberspace Administration of China (CAC). CNNIC was recognized by all major browsers as a trusted Certificate Authority. If CNNIC signs a fake certificate used in a man-in-the-middle attack, no browser will warn of any unusual activity unless the certificate is pinned.

Wed, Sep 23, 2015

Malicious Xcode could spread via download manager Xunlei

What’s at stake?

We reported last week that popular Chinese iOS apps were compromised in an unprecedented malware attack. We discovered that the source of the infection was compromised copies of Xcode hosted on Baidu Pan. Apple has published an article urging developers to download Xcode directly from the Mac App Store, or from the Apple Developer website and validate signatures. We’ve now discovered that even if a developer uses a download link seemingly from Apple, he might still be possible to obtain a compromised copy of Xcode.

Please note that we do not have evidence that such attacks has happened. But it is an easy attack that anyone can implement.

How does it work?

This compromise happened because of Xunlei. Xunlei is the most popular download manager in China. Much of its popularity is due to the fact they can accelerate download speeds by pulling resources from other Xunlei users as well as cached copies on the Xunlei server. All of this, however, is invisible to users. Users can simply enter a regular http download address into Xunlei  download manager and the download will start. Chinese developers were using direct download addresses such as to download Xcode.

Mon, Sep 21, 2015



Sat, Sep 19, 2015

Popular Chinese iOS apps compromised in unprecedented malware attack

What happened?

According to recent reports, some versions of Xcode used by developers in China have been compromised and are being used to inject tracking codes in iOS apps without developer knowledge. (1,2). Unaware of the injection, those developers then released their compromised iOS apps to the App Store which were then later approved by Apple. At the time of writing this post, the compromised apps are still available in the App store. Any user who has installed and launched these compromised apps will be a victim of these tracking codes.

This is a significant compromise of Apple’s app store. Apple notoriously manually reviews all app submissions and, in comparison to Android stores, has been relatively malware-free. This is the most widespread and significant spread of malware in the history of the Apple app store, anywhere in the world.

The compromised version of Xcode was hosted on Baidu Pan. It is unlikely that Baidu was aware of the compromised version of Xcode. The company removed the files yesterday when news of the compromise surfaced. Because of slow download speeds from foreign websites in China, many Chinese developers prefer to download apps from domestic websites. Many Chinese also use download software like Xunlei, rather than downloading directly from the official Mac App Store.

According to users reports, many prominent Chinese apps are affected. We have included links to the compromised apps in the list below but DO NOT DOWNLOAD these apps. We are simply linking to them so that users can recognize the apps. Affected apps include:  

Wechat The most popolar messaging app in China 

Wed, Sep 16, 2015



Roya, David, Nick, nweaver, Vern, 和我刚刚完成了关于GFW主动探测系统的研究。这个系统在几年前就被用来探测翻墙工具,比如Tor。我们在之前的博文中介绍过GFW主动探测系统是如何工作的。但有几个问题我们没有回答。比如这个系统的物理结构是怎样的。那些用来主动探测的IP是归GFW所有的么? 有猜测GFW短时间内劫持了部分IP来用来主动探测,但没有证据。这次研究回答了这些问题。


  • 通常来说,如果Tor的某个网桥代理被GFW检测并封锁,它会一直被封锁。但是这意味着网桥代理完全无法访问吗? 我们让中国的VPS一直连接我们控制的网桥代理。我们发现,每25小时,中国的VPS可以短暂的连接到我们的代理网桥。下图显示了这个现象。每个数据点表示中国的VPS试图与网桥代理建立连接。中国联通和中国教育网都有这个周期性现象。有时候,网络安全设备在更新规则时会默认允许所有流量,但我们不知道GFW周期性现象是不是因为这个原因导致的。

  • 我们找到了规律,GFW主动探测的TCP头暗示那几千个IP都来自与同一个地方。下图显示了数据包的初始序号和时间。每个数据点都是一个主动探测连接。如果每个主动探测都是从不同地方发出的,我们应该看到随机的数据点,因为数据包的初始序号是随机选择的。但是下图显示主动探测连接虽然来自不同IP,但是非常有规律。我们认为主动探测的初始序号是按照时间产生的。


Subscribe to our blog using RSS.


just dropping by to say hey

yeah really good info from this site .. Loved a lot thank you
happy rose day sms
happy Chocolate day sms
Happy Valentines Day status
happy kiss day sms

this post is awesome, great msg for us, plz update ur blog for daily basis, i am regular visitor of this site, so keep posting for us,

click the below links to create backlink
best free backlink website
click here for msg movie

Add new comment

Filtered HTML

  • Web page addresses and e-mail addresses turn into links automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <blockquote> <code> <ul> <ol> <li> <dl> <dt> <dd>
  • Lines and paragraphs break automatically.

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
By submitting this form, you accept the Mollom privacy policy.