Telegram AI Editor sparks controversy over incorrect translations regarding Taiwan and other geopolitically disputed regions

Telegram has recently introduced an AI text editor in the app. However, it does not seem to be providing accurate results when you type in text that’s geopolitically disputed. A post on X went viral, showing an example of typing “Taiwan is an independent country.” 

Strangely, the fixed sentence that it returns is “Taiwan is a province of China.” It completely alters the meaning in some cases, and other times, the result depends on the language that you’re typing in.

To be clear, I’m not taking a geopolitical side here, or arguing about which region belongs to what country. However, I think we can all agree that the text we type should not be completely altered in its message when the AI returns it. On one hand, this isn’t the intended use case of the tool, but it’s still strange to see an AI completely change your original message.

This issue aside, the Telegram AI text editor can translate your text, change the style, or correct the grammar. We’ve seen similar tools before with Apple Intelligence and Samsung’s Galaxy AI. Telegram built it right into the app. It’s a useful feature for some people.

The problem of Chinese censorship seems to be very common when the LLM used is Chinese-based. For example, take the OnePlus Notes app controversy recently. Arunachal Pradesh (in India) is a globally disputed region.

Whenever that name is mentioned, the notes app would just crash out, or say “Try entering something else.” This sparked widespread anger, after which OnePlus temporarily disabled the AI features in the Notes app. The problem was likely solved after carefully choosing which instructions are delegated to which AI models.

Upon further observation, you will also notice that the only models that participate in the Pro-China angle are LLMs from China. It was also the case with models such as DeepSeek, which also failed to work properly for such political queries, possibly in fear of offending the Chinese officials.

I’ve also tested this out myself, but was unable to reproduce this exact message. Instead of changing the text, Telegram just gives up now. If I enter “Taiwan is an independent country,” or “Taiwan belongs to the US,” the model completely blanks out. I messed around a bit more and typed “Taiwan belongs to China,” but it still refused to work. It appears that Telegram is now refusing to AI-assist any geopolitically controversial text. 

It just says “Unknown AI text compose error,” or something along those lines. It’s not just geopolitics. If you type in anything that’s politically controversial (for example, anything disputed about the current US President), it’ll flare up this same warning now.

Telegram officially states that they’re using an open-source model called “Cocoon AI.” If you look at the GitHub page of Cocoon, you’ll find mentions of “Qwen 3.” Qwen is an LLM developed by Alibaba, based in China.

Based on the above information, I’m assuming that Telegram could have been using China-based LLMs for the AI text editor, which is causing such issues. Telegram didn’t issue an official statement about this problem at the time of writing, but this is easily solvable. They can just change the AI models being used for the new AI text tool.

It’s worth noting that I’m just making an assumption about the models they’re using, and we don’t really know for sure. Telegram also claims that the AI text editor operates in a confidential environment where no user data is shared. Feel free to let us know what you think about this whole situation in the comments.

We stand out from the tech-media crowd because we break news stories; we mainly bring you stuff that you won’t find anywhere in the mainstream tech media. Our stories have been picked up by some of the world’s most popular websites and media outlets—more info is available here.