NebuSec Claims Telegram Android Widget Defenses Can Be Bypassed

Security firm NebuSec says it has demonstrated two ways to run code beyond Telegram’s new Android widget protections in Telegram for Android 13.2.2 (7303). In a video recorded against a Telegram test server, a bot widget bypasses the filter around the widget engine and opens the calculator application on the device.
The claim follows Telegram’s move to isolate QuickJS, the JavaScript engine used for widgets, in a separate Android process and to add a seccomp system-call filter. Widgets are interactive blocks that bots can send in chats, and their JavaScript executes on the user’s phone. General bots are currently not allowed to use widgets on the live service, including in private chats, according to a late-week comment in a telegram-bot-api GitHub thread.
Two claimed paths beyond the filter
NebuSec says one technique relies on a widget vulnerability that the seccomp policy does not cover. The second reportedly combines a widget flaw that the policy does cover with a kernel vulnerability. The researchers have withheld the technical details, named no CVE identifiers, and have not said whether the issues were reported to Telegram.
The demonstration establishes code execution beyond the stated filter, but it does not establish the broader impact. NebuSec’s material does not say what data or functions the code could access, including Telegram messages or the rest of the device. It also does not clarify whether a victim must start a bot or tap a widget before exploitation can occur.
Why the sandbox design is under scrutiny
Telegram’s widget documentation describes a sandbox that includes restrictions such as no network access. NebuSec argues that these controls reside in the engine itself. A use-after-free or out-of-bounds write in QuickJS could allow a script to read and write the engine process memory, making those in-engine rules ineffective.
The published 13.2.2 (7303) Android code names several dozen system calls in its seccomp policy, including ptrace, bpf and mount, while allowing calls not named by the policy. NebuSec notes that ioctl is not filtered. The researchers also say one engine process runs every widget, so a compromised widget from one bot would share memory with widgets from other bots.
The Hacker News found that the published filter is configured for audit mode. In that mode, attempts to launch new programs or processes are still blocked, while other listed calls are logged and allowed. A code comment says the audit configuration will remain until device logs show that the engine does not make those calls. The published code does not establish whether production builds installed by users use the same setting.
Telegram and Kaspersky statements
Pavel Durov, Telegram’s chief executive, said after earlier widget RCE reports that receiving a message alone could not compromise a user. He said a person first had to start a third-party service and that the issue had been fully patched on the server side. That statement was made before NebuSec’s demonstration.
Separately, Kaspersky reported suspected exploitation involving Telegram widgets in some channels. It said a PowerShell script ran without the user noticing before downloading and launching WebRAT, also known as Salat Stealer, which can control systems and steal data from crypto wallets and Telegram accounts. Kaspersky did not identify the underlying flaw or Telegram version, report victim numbers, or connect the activity to either public demonstration.
For businesses, the practical response is to limit use of untrusted bot widgets, maintain current Telegram client updates, and wait for confirmed vendor guidance on affected versions and remediation scope before treating widget-based workflows as suitable for sensitive activity.

