Webworm VPN Bridge, Bazuka, Fake GTA VI: 21 Nov Roundup
TL;DR · the short version
Four research publications from 18-20 November 2026. Rapid7 found a modified SoftEther VPN Bridge, rebranded NtMonitor VPN, that it attributes to the China-aligned espionage group Webworm with moderate confidence; it found no extra malware, so the VPN bridge itself is the access. ThreatFabric disclosed Bazuka, a 23-module Android fraud framework with overlays, SMS interception, NFC card relay and an LLM-assisted control panel; it saw no distribution through Google Play. Bitdefender found a fake "GTA VI cloud" PC client that uses a ClickFix PowerShell paste to install a Rust-based stealer of browser, wallet and AI API credentials; there is no official GTA VI for PC. NebuSec showed one malicious webpage could compromise Meta's Muse agent and read chats and memories, patched by Meta before publication and without escaping the hypervisor.
Four pieces of threat research landed between 18 and 20 November 2026. Rapid7 found a disguised SoftEther VPN bridge it ties to the espionage group Webworm (moderate confidence). ThreatFabric disclosed Bazuka, an Android fraud framework (no Google Play distribution seen). Bitdefender found a fake GTA VI PC client that installs a stealer through a ClickFix paste. NebuSec showed one webpage could compromise Meta's Muse agent, patched before publication. Each section below is taken from the researchers' own write-up.
Who this is for: security leaders who want the week's research in one place with the "so what" for their own estate.
All four primary sources read on 22 November 2026. PlayCISO is not affiliated with Rapid7, ThreatFabric, Bitdefender, NebuSec or Meta. No hashes, domains or IP addresses are repeated; take them from the researchers' indicator lists.
Confirmed vs not yet known (22 November 2026)
Webworm: the modified SoftEther build and its behaviour are Rapid7's analysis. The Webworm link is "moderate confidence". Not known: how the attackers got in, and which organisations were targeted.
Bazuka: capabilities are ThreatFabric's analysis of samples and an exposed panel. No Google Play distribution was seen; no known group is linked; target countries and banks are not named.
Fake GTA VI client: the chain and stealer are Bitdefender's analysis. Victim numbers are not given.
Meta Muse: the exploit chain is NebuSec's account. NebuSec says Meta patched it before publication and that no hypervisor escape was shown. Meta has not commented publicly that we found; no CVE is listed.
1. Webworm's custom SoftEther VPN bridge (Rapid7)
On 19 November Rapid7 described a modified build of the open-source SoftEther VPN Bridge 4.44 (build 9807). Its file metadata calls it "Network Monitor Software", strings are rebranded "NtMonitor VPN", the handshake watermark was swapped for a Windows-logo image, the resource archive is embedded in the executable, and the configuration, renamed setting.conf, is encrypted with a custom cipher. Every configuration Rapid7 decrypted bridges a remote network into the local one; in non-test cases the remote end sat on infrastructure in the Vultr network (AS 20473).
The important line: "we did not identify any additional malware functionality." No backdoor, no command execution. Legitimate VPN software, configured by the attacker, is the whole implant, which is why Rapid7 calls it "zero lines of malware". Rapid7 attributes it to Webworm, a China-aligned group known for targeting governments, mainly in Europe, "with moderate confidence", based on ESET's earlier reporting and the infrastructure.
What to do: list every VPN and tunnelling tool allowed in your estate and alert on anything else. Rapid7 notes the Windows service is still called SEVPNBRIDGE. Hunt for new services, new firewall rules and persistent outbound connections that match no approved product, and load Rapid7's YARA rules. The Edge VPN Intrusion Lab and the C2 Detection Lab rehearse both halves.
2. Bazuka Android fraud framework (ThreatFabric)
On 18 November ThreatFabric published Bazuka, found in late October while investigating an NFC-relay banking trojan. What first looked like a wrapper around an open-source project turned out to be "a full malware framework": a dropper or downloader, a core and 23 modules, loaded in memory. Capabilities include overlays (including a screenshot of the home or lock screen shown on top to hide what is happening underneath), keylogging, SMS and 2FA interception, remote control and screencasting, clipboard crypto-address swapping, call forwarding and an NFC module that talks to contactless bank cards. It grants itself permissions through Accessibility or ADB.
The control panel integrates local and API-hosted language models; sample prompts ask for a victim "score" and a strategy for clicking through permission prompts. Droppers posed as an adult "premium" TikTok app and IPTV apps. Google Play: the panel is built for it, but ThreatFabric says it has "not seen any signs of it being a real case". No known group is linked; The Hacker News notes the panel mixes English and Russian.
What to do: block sideloading on managed Android devices, alert when an app requests Accessibility rights it has no reason for, and remind staff that "premium" versions of popular apps are a classic lure. Your fraud team should know that SMS one-time codes and tap-to-pay are both in scope. Practise in the Android Security Lab.
3. Fake GTA VI cloud client installs a Rust stealer (Bitdefender)
On 20 November Bitdefender Labs described a site offering a "GTA VI cloud" PC client that claimed to stream the console game. GTA VI is out on PlayStation 5 and Xbox Series X|S; per Bitdefender, "no official PC version is available." Visitors could download a ZIP or follow ClickFix instructions: press Win+R and paste a PowerShell command that bypasses the execution policy and runs a remote script. The client plays the trailer as a decoy while it side-loads a DLL beside a legitimate Windows binary, which injects a Rust-based information stealer into other processes without writing it to disk.
Useful next, if this was relevant to you
The stealer takes browser passwords, cookies and tokens, Discord and Steam data, crypto wallets, AI and developer service API keys, clipboard contents and screenshots. Bitdefender gives no victim count.
What to do: this lands on corporate laptops through gamers and on developer machines that hold cloud and AI keys. Block or alert on PowerShell launched from the Run box with irm or iex, consider turning off Win+R for standard users, and tell staff that no legitimate site asks you to paste a command to install anything. If anyone ran it: isolate the PC, then revoke sessions and rotate passwords and API keys from a clean device. See our write-up of another ClickFix chain, ROMULUS.
4. Meta Muse agent compromised by one webpage (NebuSec research)
On 20 November NebuSec published how it compromised Muse, Meta's personal AI agent, and a companion architecture write-up. Per NebuSec, each user gets a small Cloud Hypervisor virtual machine running a container with the agent and a Chromium browser inside. Making the agent load an attacker's page exploited a Chromium bug for code execution in the container; a Linux kernel bug then gave root inside that user's VM. From there, NebuSec says, conversation history and stored memories could be sent to an outside server "without triggering an approval prompt", bypassing Sentinel, the component that watches outbound traffic and asks the user to approve it, and memories could be altered.
Two limits matter. NebuSec says "all exploits described in our research had been patched by the time of publication", and "We did not demonstrate an escape into the underlying hypervisor host", so other users' VMs were not shown to be reachable. This is research, not a reported attack on Muse users. It is a different issue from the Muse Mac client zero-day disclosed in September.
What to do: if your organisation is piloting browsing agents, assume the agent's browser will meet hostile pages. Ask vendors how browser sessions are isolated per user, what an approval prompt actually covers, and whether memory can be changed by content the agent reads. Test your own agents in the Agent Prompt Injection Harness Lab and rehearse containment in the VM Escape Triage Lab.
Timeline
| Date | Event | Source |
|---|---|---|
| Late Oct 2026 | ThreatFabric finds Bazuka while investigating an NFC-relay banking trojan | ThreatFabric |
| 18 Nov 2026 | ThreatFabric publishes Bazuka | ThreatFabric |
| 19 Nov 2026 | Rapid7 publishes the Webworm SoftEther bridge analysis, with YARA rules and a config decryptor | Rapid7 |
| 20 Nov 2026 | Bitdefender publishes the fake GTA VI client and stealer | Bitdefender |
| 20 Nov 2026 | NebuSec publishes its Meta Muse research (patched before publication) | NebuSec |
What security leaders should do this week
- Get an approved list of VPN and remote-access software and alert on anything outside it, including legitimate tools such as SoftEther installed where they do not belong.
- Close sideloading and Accessibility abuse on managed Android devices, and brief the fraud team on overlay, SMS-code and NFC-relay fraud.
- Put a ClickFix rule in place: detection for Run-box PowerShell with download-and-execute, plus a one-line staff message: never paste a command a website gives you.
- Inventory AI and developer API keys on laptops; stealers now look for them by name. Move them to a vault and shorten their lifetimes.
- Ask AI agent vendors three questions: how is the agent's browser isolated, what does the approval prompt really gate, and who can change the agent's memory?
More from this week: the Payload ransomware Group Policy attack and the Party Panel Android banking trojan.
FAQ
What is the Webworm SoftEther VPN bridge?
Rapid7 found a modified build of the open-source SoftEther VPN Bridge (version 4.44, build 9807), rebranded "NtMonitor VPN" and disguised as "Network Monitor Software". In every configuration Rapid7 decrypted it bridges a remote network into the victim network. Rapid7 attributes it to the China-aligned espionage group Webworm "with moderate confidence" and says it found no additional malware functionality: the VPN bridge alone is the access.
How do I detect a rogue SoftEther VPN bridge?
Rapid7 says the Windows service is still named SEVPNBRIDGE, the configuration file is setting.conf (encrypted, starting with the bytes RCEN) and the HTTP endpoints were changed to /update/alive.cgi and /update/picture.cgi. Compare installed services, new firewall rules and long-lived outbound connections with your approved VPN inventory, and use the YARA rules and indicators Rapid7 published on GitHub.
What is Bazuka Android malware?
Bazuka is a modular Android fraud framework disclosed by ThreatFabric on 18 November 2026: a dropper or downloader, a core and 23 modules for overlays, keylogging, SMS and 2FA interception, remote control, clipboard crypto-address swapping and an NFC module for contactless cards. Its control panel can use large language models to score victims. ThreatFabric found it while investigating an NFC-relay banking trojan in late October.
Is Bazuka on Google Play?
No evidence of that. ThreatFabric describes Google Play as a channel the panel is designed for but says it has "not seen any signs of it being a real case". The droppers it saw posed as an adult "premium" TikTok app and IPTV apps distributed outside the Play Store.
Is there a GTA VI PC download or cloud streaming client?
Not an official one. Bitdefender says GTA VI is out on PlayStation 5 and Xbox Series X|S and "At the time of writing, no official PC version is available." A site offering a "GTA VI cloud" PC client told visitors to paste a PowerShell command into the Run box, a ClickFix trick, which installed a Rust-based information stealer.
What does the fake GTA VI stealer take?
Per Bitdefender: saved browser passwords, cookies and tokens from Chromium and Firefox-family browsers, Discord and Steam data, cryptocurrency wallets, AI and developer service credentials and API keys, clipboard contents and screenshots. If you ran the command, treat the PC as compromised and change passwords from a different, clean device, starting with email.
Was Meta Muse hacked?
Researchers at NebuSec showed that a single malicious webpage could compromise Meta's Muse agent: a Chromium bug gave code execution in the agent's container, a Linux kernel bug gave root in that user's virtual machine, and chats and memories could be sent out "without triggering an approval prompt". NebuSec says Meta patched everything before publication and that it "did not demonstrate an escape into the underlying hypervisor host". It was research, not a reported attack on users.
Sources
- Rapid7: Spying on Governments with Zero Lines of Malware: A Custom SoftEther Bridge in Webworm's Toolkit (Milan Spinka, 19 Nov 2026)
- ThreatFabric: Bazuka, the fraud framework that can do it all (Mobile Threat Intelligence team, 18 Nov 2026)
- Bitdefender Labs: Fake GTA VI PC client spreads malware through a ClickFix attack (Janos Gergo Szeles and Silviu Stahie, 20 Nov 2026)
- NebuSec: Hacking Muse: Stealing Private Chats Through Browser Use and Inside Meta Muse: Browser, Sentinel and the Shared-Kernel Boundary (20 Nov 2026)
- The Hacker News: Bazuka Android malware uses fake screen overlays (Ravie Lakshmanan, 20 Nov 2026), secondary coverage
Educational guidance for security leaders, not a threat-intelligence report. Check each researcher's write-up for indicators before acting.