A Telegram Android voice call flaw present in every release from 8.6.2 through 12.10.3 could be chained into full control of a phone once the victim answered a call, according to security startup NebuSec. Telegram fixed the parsing bug in version 12.10.4 in late September, so the job for IT teams now is finding the phones that never received it.
Key takeaways
- NebuSec says Telegram for Android 8.6.2 through 12.10.3 contains a call-parsing flaw it chained with two unpublished bugs to take over a phone.
- The trigger is answering a one-to-one voice call from an ordinary Telegram account; declining the call or setting calls to Nobody blocks it.
- Telegram fixed the bug in Android 12.10.4 but disputes the remote code execution claim, so no CVE exists and CVE-based scanners will not flag it.
- CVE-2026-111105 is a separate, unpatched rlottie sticker flaw; Telegram for Android moved to the Rust-based tlottie by 12.10.0.
What happened in the Telegram Android voice call flaw?
NebuSec, a Y Combinator-backed security startup, disclosed on October 29, 2026 that a memory-safety bug in the calling code of Telegram for Android can serve as the first link in a three-part chain. The other two links are an information-leak bug and an Android root exploit. Together, the company says, they give an attacker the whole phone: messages, microphone and camera.
The attacker needs nothing more than an ordinary Telegram account and a victim who answers. The fix is already out: Telegram shipped Telegram for Android 12.10.4 in late September 2026, and it corrects the faulty parsing.
NebuSec reported the bug to Telegram on September 26, after that release had gone out. Telegram acknowledged the report but did not accept that the flaw could be turned into remote code execution. So there is no CVE identifier, and no bug bounty was paid.
The public evidence for the full chain is a demonstration video showing two phones, one marked attacker and one marked victim. The victim accepts a Telegram call. About 30 seconds later the phone returns to the home screen, the app icons tumble down the screen and a “HACKED BY NEBUSEC” logo appears.
The video does not show which versions the victim phone ran, and as of October 29 NebuSec had published no technical write-up, so the leak bug and root exploit exist publicly only in the company’s statements.
The parsing flaw itself stands on firmer ground. The Hacker News compared NebuSec’s description against Telegram’s published source for 12.10.3 and found a flaw that matches it.
Why did the same tgcalls fix reach Android last?
The fix was written once, in April, then shipped to each app on its own schedule, and Android came last. Telegram’s Android, iOS and desktop apps share one calling library, tgcalls. A commit labeled “Fix camera rotation” landed there on April 21, 2026. The modest name covers a memory-safety correction.
| Client | Version with fix | Released |
|---|---|---|
| tgcalls library | “Fix camera rotation” commit | April 21, 2026 |
| iOS | 12.7 | May 2026 |
| Desktop | 7.0.1 | July 2026 |
| Android | 12.10.4 | Late September 2026 |
Android users waited about five months after the commit and roughly four months longer than iPhone users. Shared source did not mean synchronized patching. Once a fix sits in a public repository but not in the shipping app, anyone reading commits can spot the change before users are protected.
What the code got wrong
During call setup and throughout a call, the two phones exchange short MediaState messages. These messages report simple status flags: whether the microphone is muted, whether the camera is on, and how the video is rotated.
In 12.10.3, when a MediaState message carried a rotation value, the parser checked that value and then read a different field, the video state, whether or not the message contained it. The code did not handle the missing field and touched memory it should not have. Version 12.10.4 reads the rotation from the correct field.
It is the kind of mistake that survives code review, because a check sits right there and looks correct. But this parser handles data sent by the other caller, data the person answering has no control over.
How it was found
NebuSec says its Code Security product found the flaw, and that the product runs on OpenAI’s GPT-5.4 model. This is not the company’s first Android chain. In July 2026, NebuSec described one that, by its account, started with a Firefox flaw and reached the kernel of an ARM64 device running Android 17.
A check on one field followed by a read of another is exactly the kind of mismatch automated code review is good at catching, and that tired human reviewers miss.
Can answering a Telegram call really hack your phone?
According to NebuSec, yes, on a vulnerable Android build where the other two bugs in the chain also work, and only if the victim answers. Eten Zou, NebuSec’s co-founder and chief executive, put it plainly: the attack needs the victim to be called and to answer. Declining the call blocks it. That makes this a one-click attack, not a zero-click one.
NebuSec has not seen the flaw used in real attacks and believes nobody else had realized it could be turned into remote code execution. No attacker, victim or indicator of compromise has been named.
Call code is still an attractive target. It parses data from a stranger the moment a call connects, before the user has done anything beyond tapping Answer, and it runs on a very large number of phones.
Which versions are affected
NebuSec says every Telegram for Android build from 8.6.2 (March 2022) through 12.10.3 (September 2026) can trigger the flaw, roughly four and a half years of releases. The start date has a code-level reason: 8.6.2 is the release that switched on the calling path that processes MediaState messages.
The confirmed scope is narrower than “all Telegram users”:
- Platform: Telegram for Android only. NebuSec has no evidence that the iOS or desktop apps are directly affected.
- Call type: one-to-one private voice calls. Nothing shows group calls are affected.
- Hardware: the chain was confirmed on ARM64 phones, the architecture most current Android handsets use. The Android versions used in testing were not disclosed.
Because the bad code lived in tgcalls, the same mistake existed in the iOS and desktop source until each received the fix. Shared code is not proof of shared exploitability, though. Memory layout, compiler settings and platform defenses differ, and nobody has shown those builds could be attacked this way.
Treat the Android window as the confirmed risk, and the other clients as patched and unproven.
What should a business do if staff use Telegram on Android?
Get every Android phone that runs Telegram onto 12.10.4 or later. Verify the installed build rather than assuming it, and tighten call privacy while you work through the list. Here is the routine to run this week.
1. Find the build on each phone
- In the app: the version number appears at the bottom of Telegram’s Settings page. It should read 12.10.4 or higher.
- In Android: Settings, Apps, Telegram, App details shows the same number. Install any pending update from the legitimate source that phone normally uses.
- On an authorized device with USB debugging enabled, an admin can run adb shell dumpsys package org.telegram.messenger and check that versionName is 12.10.4 or later. The build downloaded from Telegram’s website installs as org.telegram.messenger.web, so check both package names.
- If you manage phones through MDM, pull the app inventory and filter for Telegram builds below 12.10.4.
2. Hunt for the phones that do not auto-update
In our mobile and BYOD reviews, the phones that miss updates like this are rarely the ones IT expects. They are the APK a manager sideloaded from Telegram’s website years ago, the company phone with automatic updates switched off to save data, and the managed store pinned to an approved version.
Then there is the spare handset in a drawer, still signed in to the sales team’s chats. A Play Store check alone misses sideloaded builds. Read the version on the device, not the store listing.
3. Restrict who can call
In Telegram, open Settings, Privacy and Security, Calls, and set “Who can call me” to Nobody, or to My Contacts if staff need calls for work. Review the exceptions list too, because an exception can still let a specific account through.
Since the trigger is an answered call, the Nobody setting blocks the path NebuSec demonstrated, and so does declining calls from unknown accounts. My Contacts narrows exposure but still trusts every saved contact. Treat either setting as a stopgap for phones you cannot update today, not a substitute for the update.
4. Check Android updates separately
Updating Telegram fixes the first link and nothing else. App updates and Android security updates are independent, so 12.10.4 cannot patch the unnamed root exploit later in the chain. Check Settings, then Security and privacy or System, then Security update, or read the patch level with adb shell getprop ro.build.version.security_patch.
Nobody can say which patch level defeats NebuSec’s root exploit, because it has not been published. Phones that no longer receive Android security updates deserve a hard look regardless, and if they handle company data they are candidates for replacement.
A vulnerability management program that tracks app and OS versions together catches this kind of gap without a scramble.
What if someone answered a suspicious call on an old build?
Review the phone and the accounts signed in on it, and do not let a clean antivirus scan settle the question. There are no public indicators of compromise or traffic signatures for this chain, so call history can neither confirm nor rule out an attack.
An answered call from an unknown account on a build older than 12.10.4 is a reason to look, not proof of compromise. Do not wait for falling icons either. The animation and logo in NebuSec’s demo are effects chosen for the camera. A real attacker with root access would stay quiet.
If the full chain had run, updating or uninstalling Telegram would not remove root-level persistence. Work from that assumption:
- Take the device off corporate Wi-Fi and VPN, and preserve what you can: call log, installed apps, MDM records.
- From a known-clean device, rotate passwords and end sessions for accounts used on the phone, starting with email, single sign-on and Telegram itself (Settings, Devices, terminate other sessions).
- For high-risk users such as finance staff, executives or anyone who approves payments, follow your device rebuild or replacement process instead of trusting the existing install.
If the phone belonged to someone whose mailbox or Telegram account could approve payments or impersonate the company, account takeover recovery covers session revocation and the follow-up checks.
If the worry is broader, that an attacker already has a foothold in the network, a threat hunting and compromise assessment gives an evidence-based answer.
Why do Telegram and NebuSec disagree on remote code execution?
Telegram does not accept that the bug can become remote code execution, and NebuSec has not yet published the details that would settle the question. That disagreement is why there is no CVE and no bounty. The Hacker News said it had asked Telegram for comment.
Both positions can be partly right. A bad memory read on its own is often not enough to run code. In NebuSec’s account, the call flaw plus the leak bug gets the attacker out of the Telegram app and onto Android, and the root exploit then takes the whole phone. That is exactly the shape of the chain attackers usually need.
Telegram may be judging the single flaw while NebuSec claims the chain. Until the information-leak bug and the root exploit are documented, outsiders cannot verify the full claim.
For a business, the dispute changes nothing. A CVE is an identifier, not an independent ruling on exploitability. Scanners and patch dashboards that key on CVE IDs will not flag this flaw, so check Telegram versions directly. Waiting for a CVE before installing a fix that already exists means waiting for paperwork.
What is still unknown
- The information-leak bug and the Android root exploit: not named, not published, not independently checked.
- Which Android versions and security patch levels the chain works against, and whether non-ARM64 phones are affected.
- Whether anyone used the call flaw before NebuSec’s report. No victims, actors or indicators have been disclosed.
- Whether the pre-fix tgcalls code was exploitable in the iOS or desktop clients.
- Telegram’s public position, beyond what it told NebuSec.
What about the separate rlottie sticker bug, CVE-2026-111105?
CVE-2026-111105 is a different bug in a different component, with a different status. The Zero Day Initiative (ZDI) published it on October 15, 2026 as an unpatched flaw in rlottie, the library Telegram apps have used to draw animated stickers. ZDI scores it CVSS 7.0 and says it can let an attacker run code if a user opens a malicious file or page.
That score belongs only to the rlottie issue; the call flaw has none. Telegram told ZDI in March 2026 that checks on its servers mitigated the flaw, and in April asked ZDI for a full remote code execution chain. Publication by itself does not mean anyone has exploited it.
Researcher Saifeldeen Aziz says those checks could be bypassed, according to the abstract for Aziz’s OffensiveCon talk in Tokyo. The research used AI-guided fuzzing, which feeds code large amounts of broken or unexpected input, to find memory-corruption bugs in rlottie across Telegram’s iOS, Android and desktop apps.
It then chained those bugs with flaws in how stickers are delivered and checked to run code on Telegram for iOS.
Telegram’s practical answer has been to replace the library with tlottie, written in Rust, according to its published source code:
- Android: switched between 12.8.1 (June 2026) and 12.10.0 (August 2026)
- Desktop: switched in September 2026
- iOS: added tlottie in 13.0 (October 2026), but still includes rlottie
Safe Rust rules out whole classes of the memory errors behind bugs like these, though unsafe blocks, interfaces to C code and plain logic errors can still produce flaws. For Android fleets the news is simple: a phone on 12.10.4 is already past both the call fix and the sticker library swap.
If NebuSec publishes its write-up, the patch-level questions may get answers. Until then, the Telegram version on each phone is the one thing you can verify today, and 12.10.4 is the number to look for.
FAQ
What Telegram Android version fixes the voice call bug?
Telegram for Android 12.10.4, released in late September 2026, fixes the MediaState parsing flaw. Any build from 8.6.2 through 12.10.3 is in the affected range NebuSec describes. The version number appears at the bottom of Telegram's Settings page.
Can answering a Telegram call hack your phone?
NebuSec says it can on a vulnerable Android build when the call bug is chained with an information-leak bug and an Android root exploit. The victim must answer; declining the call blocks the attack.
Does the Telegram call bug affect iPhone or desktop?
NebuSec has no evidence it directly affects iOS, desktop or group calls. The shared tgcalls fix reached iOS in 12.7 and desktop in 7.0.1, and exploitability on those clients has not been shown.
Was a CVE assigned to the Telegram Android call flaw?
No. Telegram acknowledged the report but did not accept that the flaw could become remote code execution, so no CVE was assigned and no bounty was paid. The fix exists regardless.
How do I block unknown Telegram voice calls?
Open Settings, Privacy and Security, Calls, and set "Who can call me" to Nobody or My Contacts, then review the exceptions list. Use this as a stopgap until the app is updated.
What is CVE-2026-111105 in Telegram?
It is a separate CVSS 7.0 flaw in rlottie, the library Telegram apps have used for animated stickers, published unpatched by the Zero Day Initiative on October 15, 2026. Telegram has been replacing rlottie with the Rust-based tlottie.
If a staff phone answered an unknown Telegram call on an old build and you want a second opinion before rebuilding it, our incident recovery team can review the device and the accounts signed in on it, and tell you whether a rebuild is warranted.
Tell us what you're running
Scoping is free. We reply within one business day, and under 30 minutes for active incidents.
