Chinese Censors Demonstrate Ability to Block Images in Instant Messaging

Chinese users of the messaging app WeChat are now subject to levels of censorship never seen before — images sent in a one-to-one conversation can be blocked mid-transit, presumably by an algorithm, ensuring the receiver never sees what was sent to them.

Originally, software was only in place to detect certain words that would be politically objectionable to the Chinese government. To get around this, users turned to sending information via images. However, it seems even these are now not safe to send without being blocked.

Chinese dissident Liu Xiaobo passed away on Thursday; in the wake of his death, the one-to-one image restrictions seem to have been implemented. Wu Yangwei, a friend of Liu, said he attempted to use WeChat to send a photo of him hugging his wife before he died. He was certain he had sent the photo, but his friends never received it. This was confirmed by the University of Toronto’s Citizen Lab, who issued a new report on the subject.

Citizen Lab posited that due to the speed of the interception, WeChat must be filtering images using an algorithm. Researcher Lotus Ruan explained that while they had noticed general censoring of images before, they had not previously observed person-to-person, targeted image blocking when they investigated Chinese censorship only a few months ago. Tencent Holdings Ltd., the corporation responsible for WeChat, refused a request for comment from the Wall Street Journal.

Researchers noticed that 19 images, all related to Liu, were blocked in one-to-one chats, including a cartoon of an empty chair, which is a reference to the ceremony where he won the Nobel Peace Prize in 2010, where he was represented by an empty chair. Tests by the Wall Street Journal on WeChat seemed to confirm these findings.

This push for deeper censorship seems to be part of a more concerted effort from the Chinese government to monitor their citizens. Facial recognition software is becoming common in cities, used to catch jaywalkers and criminal suspects. “Social credit” systems are being rolled out by local governments to track the online lives of people, such as internet history and payments. Further use of algorithms is pulling back the advantage from those who wish to say things that are forbidden in China, in that the sheer volume of messages was too much for human censors to handle. Bao Pu, a Hong Kong-based book publisher that specializes in literature that is banned on the mainland, explained the situation:

If you hire a million network police, it still wouldn’t be enough to filter 1.4 billion people’s messages. But if you have a machine doing it, it can instantly block everything. It doesn’t matter if it’s a billion messages or 10 billion.

Like many automated censorship attempts, users seem to have already found ways to keep the information flowing. Jeffrey Knockel, also from Citizen Lab, said that slight changes to an image, such as its color palette, or the general metadata, will allow it to pass through the system. Mr. Wu, better known by his pen-name, Ye Du, noted that he sometimes had success by rotating the image. This arms race of new censorship technology and exploits by dissidents will likely continue in China for the foreseeable future.