Key facts
- Bazuka uses a novel overlay technique that captures an image of the home or lock screen and shows it as an overlay whenever commanded, hiding activity running behind the screen.
- Bazuka misuses Android Debug Bridge shell commands together with accessibility services so it can request permissions and approve them without user action.
- The core can open system settings, enable Developer options, enable Wireless Debugging, intercept the PIN, and establish a TLS connection to the ADB daemon.
- ThreatFabric stated that ADB bridging is turning into a common feature among modern banking trojans, letting malware gain privileges secretly and install standalone modules that remain after the host app is removed.
Dutch mobile security company ThreatFabric disclosed Bazuka, an undocumented Android malware framework, on Nov. 20, 2026. The company said it found the framework in late October 2026 while investigating an NFC-relay banking Trojan.
Bazuka is a full malware framework built from a dropper or downloader, a core, and 23 modules. The modules cover overlays, keylogging, two-factor authentication and SMS interception, remote control, and swapping crypto addresses.
Experts anticipate dropper delivery via no fewer than two paths: affiliate-style traffic brokers buying social media advertisements or operating deceptive websites, plus placement on the Google Play Store. One dropper masquerades as a premium, adult-friendly version of TikTok with package name com.tiktok.premium. Two attack-chain variants were observed: one seeks authorization to install apps from third-party sources before downloading standalone APK com.tiktok.premium.pro as a stager; the other retrieves the main payload directly with a downloader.
Bazuka uses a novel overlay technique that captures an image of the home or lock screen and shows it as an overlay whenever commanded, hiding activity running behind the screen. After loading, the framework acts as both a keylogger and a traditional overlay trojan, watching what the victim types and showing an overlay when needed on command from C2.
Bazuka misuses Android Debug Bridge shell commands together with accessibility services so it can request permissions and approve them without user action. The core can open system settings, enable Developer options, enable Wireless Debugging, intercept the PIN, and establish a TLS connection to the ADB daemon.
Communication with C2 uses two WebSocket channels, Control for issuing commands and Data for exfiltrating data and screen capture streams. The control panel integrates large language models so operators can identify victims of interest and manage infections at scale.
The Bazuka LLM panel works with models hosted locally and with models reached through an API. A sample prompt mentions a score, suggesting the AI can rate victims by how attractive they are as targets. After opening the Control socket, the bot sends device information including ThreatLevel, letting operators assess victim attractiveness within seconds.
ThreatFabric said LLM chats directly within the control panel are concerning because AI agents identify victims of interest. The company also stated that ADB bridging is turning into a common feature among modern banking trojans, letting malware gain privileges secretly and install standalone modules that remain after the host app’s removal.
Defenders should watch for automated attempts to turn on developer settings and unexpected ADB pairings as signs of Bazuka. Overlay detection stays essential, and mobile threat defense tooling must keep flagging software that renders on top of sensitive banking and payment applications. Security teams ought to include AI-assisted operations when building threat models for mobile fraud.
The core can disable Google Play protection through the accessibility services API. Bazuka contains a self-destruct routine that wipes local storage, removes the database, and shuts down the app entirely. Modules load straight from a byte array instead of writing the DEX file to storage first, and the framework’s modules and dropper lacked obfuscation even though the control panel mentioned hardening.
The panel can create accounts for traffers, cybercriminals who drive victims to malware using lures, and can build droppers with options to choose permissions specified in the Android manifest file, likely to bypass Google Play app review. The control panel displays user interface elements in a mix of English and Russian, while the core searches for interface elements in Russian, English, Chinese, and Japanese. Images and videos produced by AI fill the control panel.
ThreatFabric listed modules including call_forwarding, clip_monitor, contacts, device_control, file_manager, input_recorder, location, media, oem_power, optics, preflight, proxy, push, screen_control, self_adb, session_ui, smsmessage, system_checker, tapread-nfc_module, toast, twofa, uninstall_shield, and ussd. No known threat actor or group has been linked to Bazuka. Researchers Rajat Goyal and Fernando Ortega said, “What initially began as a targeted spyware operation aimed at users of 11 specific cryptocurrency apps”
Bazuka samples include a dropper with SHA256 hash f046b50166c8229b49839eeebcc7fd14bf3f198e4ec1bfde8a5b271e1407dec5 and a downloader with SHA256 hash 4d572669d6000304a692874c1583021dfdb4f2f113ea487e470f7e4c486e503a. Bazuka C2 traffic is encrypted with ChaCha20-Poly1305, with keys rotated, kept locally in SharedPreferences, and transmitted to C2 separately as telemetry.
Hardcoded server addresses pointed to Cloudflare subdomains of brokerok[.]org, including c2ws.brokerok[.]org, c2data.brokerok[.]org, c2proxy.brokerok[.]org, and c2api.brokerok[.]org. Servers included 95.181.213.21 in Moscow, already offline, and 170.168.61.117 in Amsterdam, operational. The control panel frontend is accessible without authentication, letting anyone examine bot operator capabilities.
Zimperium zLabs described ArcticRAT last week. That Android malware supports 129 distinct commands and uses LLM APIs in its data-exfiltration pipeline to gather monetary intelligence from intercepted SMS messages and application data. ArcticRAT infection chains use malicious sites impersonating Google Play app listings, with users redirected through malvertising, smishing, or third-party forums.
ArcticRAT abuses Android accessibility services for full remote control, uses TCP sockets for C2 communication, and asks for VPN service privileges so it can null-route network communication, which blocks Google Play Protect telemetry while staging occurs. ArcticRAT employs a Black Screen overlay that draws a black un-dismissible view and drops brightness to zero, making the device appear locked or powered off. Recent versions abuse native Wireless Debugging to initiate privileged shell sessions and execute commands.
Zimperium zLabs analyzed RatHat on Sept. 17, 2026 and believes it is linked to threat actors from China after finding LLM prompts written in Chinese. RatHat spreads via malvertising, SMS messages, and phishing pages that push APK installs from sources other than Google Play. It turns on Developer Options and Wireless Debugging so it can run at the shell level locally, with no computer required, and shows HTML overlays aimed at specific banking and cryptocurrency apps to steal account credentials. RatHat uses an AI-powered engine that processes the Android Accessibility tree as XML and sends it to an unnamed assistant that provides navigation instructions.
Zimperium identified four parallel Android banking campaigns—RecruitRat, SaferRat, Astrinox, and Massiv—reaching more than 800 apps in banking, cryptocurrencies, and social platforms on April 20, 2026. These campaigns abuse the Android overlay feature to monitor screen content, capture PIN entries, and deploy fake screens, and they use the Native Session Installation API to bypass strict sideloading rules. SaferRat manipulates system navigation to resist uninstallation, while RecruitRat applies transparency effects to vanish from the app drawer. A Google spokesperson said on Sept. 21 that no apps containing RatHat are found on Google Play and Android users are automatically protected by Google Play Protect.

Join the discussion