Security researchers at NebuSec have reported that a malicious widget is capable of bypassing security filters recently introduced by Telegram in its Android application, allowing arbitrary code to be executed outside of the intended sandbox boundaries. The discovery highlights ongoing architectural challenges in securing interactive chat elements and comes on the heels of earlier viral demonstrations and active exploitation warnings from cybersecurity firms.
Interactive blocks known as widgets are sent inside chat messages by automated bots, running JavaScript directly on the end user’s mobile device. Because general-purpose bots are currently restricted from deploying widgets on the live service, NebuSec researchers conducted their technical demonstration in a controlled environment on a Telegram test server.
The security scrutiny surrounding Telegram widgets intensified after an account named V12 posted a video on social media platform X, asserting that ordinary Telegram messages could be leveraged to execute remote code. This demonstration prompted an immediate hardening response from Telegram. In version 13.2.2 (7303) of its Android application, the platform integrated QuickJS—a lightweight engine designed to execute widget JavaScript—into a dedicated, isolated process. Furthermore, developers implemented a system call filter utilizing seccomp to record or restrict system operations executed by this process.
Despite these hardening measures, NebuSec detailed in a technical analysis that their team successfully bypassed the security filter to execute code outside the intended environment via two distinct vectors. One vector relies on an unaddressed widget flaw not covered by the filter, while the second method chains a covered widget vulnerability alongside an underlying kernel bug. In a video demonstration shared by the security firm, a bot-delivered Telegram widget running on Android 13.2.2 (7303) successfully circumvents the filter to launch the device’s calculator application.

While the public proof-of-concept demonstrated calculator execution, the researchers’ initial write-up and social media posts did not elaborate on whether the code could access sensitive areas such as local user messages, storage, or other device functionalities. Additionally, details regarding user interaction—specifically whether a user had to interact with a bot or explicitly tap a widget to trigger the vulnerability—were not disclosed. The researchers have withheld comprehensive vulnerability specifics, omitted any assigned CVE identifiers, and did not publicly confirm whether the issues had been formally reported to Telegram’s development team prior to publication.
According to official Telegram documentation, widgets are engineered to operate within strict sandboxed parameters, prohibiting network access and other potentially dangerous capabilities. Security analysts note that these fundamental restrictions are enforced within the QuickJS engine’s own codebase. If an attacker leverages a memory corruption vulnerability, such as a use-after-free or out-of-bounds write flaw within QuickJS, the script gains the ability to read and write directly to the process’s memory space, effectively neutralizing the built-in sandbox guardrails.
The newly introduced seccomp filter selectively restricts a few dozen critical system calls—such as ptrace, bpf, and mount—while allowing all other operating system calls to pass through unchecked. Source code analysis of the filter reveals that certain calls, including ioctl, remain unfiltered. Furthermore, because all incoming widgets share a single execution process, a compromised widget originating from a single malicious bot shares the same memory space as widgets belonging to any other active bot on the device.
Code inspection of the published Telegram for Android 13.2.2 (7303) source repository also revealed that the seccomp filter was configured to operate in an audit mode. In this state, the filter actively blocks attempts to spawn new programs or processes but merely logs the remaining listed system calls without terminating them. An accompanying code comment noted that audit mode would remain active until diagnostic logs indicated that the engine no longer invoked those specific calls. However, it remains unconfirmed whether production versions installed by end users enforce the exact same audit configuration.
Discussions among developers on GitHub indicate that platform-level restrictions remain tight. A contributor to the official telegram-bot-api project clarified that general bots continue to be prohibited from utilizing widgets, even within private messaging contexts. As of this week, no subsequent updates have suggested that these platform-wide limitations have been lifted.

Addressing earlier public reports of code execution vulnerabilities affecting both the Android and Windows applications, Telegram Chief Executive Pavel Durov stated that users could not be compromised simply by receiving a message. Durov emphasized that successful exploitation required a user to explicitly initiate a third-party service, adding that the specific vector highlighted in initial reports had been fully mitigated on the server side. Company representatives reiterated that issues identified in earlier demonstrations had been swiftly patched and rolled out to all users.
Concurrently, independent threat intelligence warnings have emerged from cybersecurity firm Kaspersky, which reported detecting suspected exploitation attempts targeting Telegram widget vulnerabilities during user visits to specific public channels. According to Kaspersky’s findings, the observed activity involved the silent execution of a PowerShell script that subsequently downloaded and deployed a payload known as WebRAT, or Salat Stealer. This infostealer malware provides malicious actors with remote control over compromised systems, allowing the theft of cryptocurrency wallet data and private Telegram account credentials.
While Kaspersky’s advisory did not explicitly link the observed attacks to the researcher demonstrations, nor did it disclose specific vulnerability identifiers, affected application versions, or total victim counts, the firm strongly advised users to exercise caution and avoid interacting with widgets until comprehensive vendor patches are widely distributed.
