Analysis by Canada's CitizenLab revealed that WeChat (known in China as Weixin) employs real-time automatic censorship of images shared in private chats, impacting communication in China.

WeChat was discovered to use a dual approach for image censorship. It maintains a hash index of MD5 hashes for images that have been previously banned. When a user sends an image, the system checks this index. 

If a match is found, the image is blocked before reaching the recipient, without notifying either party. If the image is not in the index, it undergoes further scrutiny using optical character recognition (OCR) to detect any sensitive text and visual similarity checks against a blacklist of images. Users are unknowingly contributing to the database of blacklisted images, according to the report.

The censorship predominantly focuses on political content, including images related to government actions and social movements. The analysis indicated that WeChat's filtering is reactive to current events, censoring images associated with significant occurrences like the arrest of Huawei's Chief Financial Officer and the Sino-US Trade War.

Different chat features on WeChat experience varying levels of censorship. Group chats and WeChat Moments are subjected to stricter filtering compared to one-on-one chats - a discrepancy that suggests that the platform prioritises monitoring public discussions over private conversations, although the latter are still not immune to censorship.

The findings highlight the opaque nature of Tencent's censorship practices, as users are not informed when their content is filtered. 

The lack of transparency raises concerns about the extent of surveillance on both domestic and international users, as content shared by non-China registered accounts may also contribute to the censorship algorithms used against users within mainland China.