In the debugging menu of the Telegram iOS 13.3 beta version (build 37388), there is a feature called “Companion”: a 3D duck will appear in the private chat window.

At present, it does nothing at all—it doesn’t reply to messages, doesn’t engage in any interaction; it just stays there, watching the conversations unfold.

There’s not much to write about a duck that doesn’t talk; what’s worth mentioning is the potential implications of such a feature. On November 9th, Telegram released Bot API 10.4, which includes new Widgets that allow robots to embed HTML, CSS, and JavaScript into chat messages, thereby creating interactive interfaces. Putting these two developments together, it seems that Telegram is paving the way for “intelligents” that can operate within private conversations: they will be able to understand the context of the discussion and directly display interactive interfaces within the messages themselves.

However, problems arose with the second aspect almost immediately. Within just a few days after the Widgets were launched, users began reporting that the client-side sandbox contained vulnerabilities that allowed remote code execution. The security team warned people not to click on unknown links from unidentified sources and advised against rushing to upgrade the software at that time.

An AI assistant that is capable of reading private chat messages, combined with a scripting environment that has just been proven to have security vulnerabilities – before this system can be officially launched, Telegram must first clarify the issues related to permissions and sandboxing mechanisms.