Kenya: Censorship-free internet?

Over the last five months we ran OONI tests in Kenya almost every day to examine whether internet censorship events were occurring in the country. Hundreds of thousands of network measurements were collected and analyzed. 1,357 URLs were tested for censorship, including both international websites and sites that are more relevant to Kenya (e.g. local news outlets). Yet, after five months of intensive testing from four local vantage points in Kenya, we found almost no signs of internet censorship in the country.

Methodology

As part of our study, we started off by updating the Citizen Lab’s Kenyan test list with our local partner, Strathmore University’s Centre for Intellectual Property and Information Technology Law (CIPIT). The aim of this was to ensure that we would be testing as many URLs that are relevant to Kenya as possible and which the government could potentially have an incentive to block. The types of URLs that were added to the Kenyan list fall under 30 categories, ranging from news media, file-sharing and culture, to provocative or objectionable categories, like pornography, political criticism, and hate speech.

All of the URLs in both the Kenyan list and the “global list” (containing internationally relevant websites) were tested for censorship through OONI’s Web Connectivity test. This test is designed to examine whether access to websites is blocked through DNS tampering, TCP connection RST/IP blocking, or by a transparent HTTP proxy.

But our testing wasn’t limited to the blocking of websites. We also tested the reachability of the Tor anonymity network through OONI’s relevant software test, and we ran other OONI tests (HTTP invalid request line and HTTP header field manipulation) in an attempt to examine whether systems (“middle boxes”) that could be responsible for censorship and/or surveillance were present in the tested network.

OONI software tests were run from four local vantage points (AS36866, AS15399, AS33771, AS36914) in Kenya. The testing period started on 26th July 2016 and concluded on 14th December 2016. Once the testing period ended, we analyzed the collected data with the aim of examining whether access to sites and services was blocked, and whether proxy technologies were present in the tested network.

Findings

Upon analysis of the network measurements collected from Kenya, we found almost no signs of internet censorship.

Out of all the URLs that we tested, the only site that was likely blocked in Kenya according to our testing is http://www.sportingbet.com. The table below illustrates that most of our attempts to establish a TCP connection to this site failed, indicating the possibility of TCP/IP blocking.

The inaccessibility of http://www.sportingbet.com was also confirmed by our local partner in Kenya. Given though that this site was only one of hundreds of gambling sites that we tested, it remains unclear why there would be motivation to block access to this particular site, and not others. It’s interesting to see though that access to the same site appears to be interfered with in other, neighbouring African countries as well.

While other measurements presented similar anomalies, we excluded them from our findings as false positives because (a) the sites in question presented high failure rates from control vantage points, and (b) our local partner was able to confirm the accessibility of the sites in question through a normal browser in Kenya.

The Tor network appeared to be accessible throughout the duration of our testing, and we did not detect any systems (“middle boxes”) that could be responsible for censorship and/or surveillance. However, this does not necessarily mean that Kenyan ISPs are not using such systems, but rather that we were not able to highlight their presence through the tests run in the specific local networks (AS36866, AS15399, AS33771, AS36914) in Kenya.

The findings of this study are corroborated by our Kenyan partners who argue that they never came across a block page or similar forms of censorship in the country over the last five months. Last year however, a site mocking the President’s frequent trips abroad was taken down due to pressure from the government, according to critics.

Acknowledgement of Limitations

While it’s positive that almost no signs of internet censorship were detected as part of this study, it’s important to highlight that these findings present various limitations.

First, this study was limited to four local vantage points in Kenya (AS36866, AS15399, AS33771, AS36914) and therefore does not include measurements from other networks, where censorship events (including power cuts as indirect forms of censorship) might have occurred. As such, the findings of this study are not necessarily representative for Kenya on a countrywide level.

Second, the testing conducted under this study was limited to a set of URLs and services. While a total of 1,357 different URLs were tested for censorship as part of this study, not all the URLs on the internet were tested, indicating the possibility that other sites and services not included in test lists might have been blocked.

Third, the testing period was limited to nearly five months, starting on 26th July 2016 and concluding on 14th December 2016. Therefore, censorship events that may have occurred in the same network before and/or after the testing period are not taken into account as part of this study.

Finally, the testing was limited to software tests that are designed to examine very specific types of censorship (as explained in the methodology section above). Other types of censorship that are not examined as part of OONI’s software tests are not analyzed as part of this study.

Contribute to this study

Even when a country appears to have no censorship, it’s important to monitor internet censorship nonetheless across time. Laws and policies change, and decisions around internet censorship can change along with them.

OONI’s software (called OONI Probe) allows users around the world to automatically test their networks every day for signs of internet censorship. When and if internet censorship occurs, network data demonstrating how it’s implemented will be collected, analyzed and published. This can help increase transparency around internet censorship.

In light of the limitations outlined above, we encourage individuals in Kenya to contribute to this study through the following ways:

1. If you’re a Linux or macOS user, consider running OONI Probe (especially if you can cover different vantage points!). Learn how to do so here.

2. If you’re a Windows user or are interested in running OONI Probe from a separate device, consider running OONI Probe’s distribution for Raspberry Pis. Learn how to do so here.

3. If you have insights on which URLs to test for censorship, consider contributing to the Kenyan test list. Find information on how to do so here.

Questions? Contact us at contact@openobservatory.org.