Censorship Found in Telegram's Built-in AI Message Editor

The neural network in the messenger is based on the Chinese Qwen3 model with content filtering typical of the PRC

Telegram has introduced a built-in AI message editor that helps users edit and improve texts. However, shortly after launch, users noticed an unusual feature: the neural network refuses to process certain topics and blocks requests that may be deemed undesirable. Moreover, the filtering works according to the rules typical of Chinese AI services.

Image source Sora

The reason, presumably, is that Telegram's AI editor is based on the Chinese Qwen3 model. This model, developed by Alibaba, by default includes censorship mechanisms in accordance with PRC law. China has strict requirements for neural networks: they must be trained only on "politically safe" data and reject requests that could lead to "undermining the state system." It is almost impossible to get rid of built-in censorship without changing the model itself.

Thus, millions of Telegram users around the world, including Russia, have gained access to an AI tool with restrictions that are set not by the messenger developers, but by the Chinese government.

Читайте ещё материалы по теме:

Sources:
Rozetked
2026-04-04 10:21:57 Olesya Mihaylova Social networks and Messengers Telegram
Telegram Дзен

Now on home

All news