Bazuka Android Malware Framework Integrates LLMs for Victim Scoring and Deploys Novel Screenshot Overlays to Conceal Theft

Cybersecurity researchers at ThreatFabric have uncovered Bazuka, a previously unknown Android malware framework that combines large language model integration with advanced overlay techniques for large-scale credential theft and device control. The framework includes a dropper, core component, and 23 specialized modules covering overlays, keylogging, 2FA interception, SMS theft, remote ADB control, and cryptocurrency address swapping. A distinctive feature allows Bazuka to capture screenshots of the home or lock screen and display them as overlays to mask malicious background operations. The control panel incorporates AI models, both local and API-based, to help operators score and prioritize high-value victims. Bazuka has been observed distributed through social engineering lures and potentially Google Play Store droppers masquerading as premium TikTok applications. Similar LLM-enhanced capabilities appear in the ArcticRAT malware detailed by Zimperium, which uses AI to parse stolen SMS and app data for financial intelligence.

The Hacker NewsTranslated from English

Read at source

Cybersecurity researchers have disclosed details of a previously unknown Android malware framework called Bazuka that offers large language model (LLM) integration for victim management and control at scale.

Dutch mobile security company ThreatFabric said it discovered the threat in late October 2026 during an investigation into an NFC-relay banking Trojan. "Bazuka is a full malware framework," ThreatFabric said in a technical report. "A dropper/downloader, a core, and 23 modules cover overlays, keylogging, 2FA and SMS interception, remote control, and crypto-address swapping."

One notable aspect of the malware that distinguishes it from the rest is a novel overlay technique that allows the trojan to take a screenshot of the home screen or lock screen, and display that as an overlay on demand to cover up malicious activities taking place in the background.

Another perhaps not-so-surprising element is the use of artificial intelligence (AI) in the control panel to help operators identify victims of interest and manage infections. It supports both locally hosted models and models accessed through an API. One of the sample prompts provided by the author refers to a score, indicating an operator can likely request the AI model to score victims and determine how attractive they are as targets.

The panel also supports the following capabilities:

  • Create accounts for traffers, who are cybercriminals responsible for driving victims to malware using various lures
  • Build droppers containing the malware with options to choose the kind of permissions specified in the Android manifest file likely in an attempt to bypass Google Play app review process

This indicates that the dropper apps can be distributed via at least two distribution channels: through traffers, who can use social media ads or bogus sites to trick users into downloading them, and Google Play Store. One of the droppers containing the malware masquerades as a premium, adult-friendly version of TikTok (package name: "com.tiktok.premium").

Two different variants of the Bazuka attack chain have been observed:

  • A dropper, which asks the victim for permissions to install applications from third-party sources and then proceeds to download a standalone APK ("com.tiktok.premium.pro") that serves as the stager
  • A downloader that directly downloads the main payload

It's suspected that the stager is the same as the downloader, whose main responsibility is to retrieve the core and supplementary modules necessary for a particular campaign, and then launch them.

"Once the core and required modules are loaded, Bazuka operates as a keylogger and a classic overlay trojan, monitoring the text entered by the victim and displaying an overlay at the right moment," ThreatFabric said. "All functionality is controlled by commands from C2 [command-and-control]."

The core module is the orchestrator, maintaining communication with C2, loading additional DEX modules, dispatching commands, collecting device information, and managing permissions and component updates. Like other Android trojans, it abuses Android Debug Bridge (ADB) shell commands and the operating system's accessibility services to request permissions and grant them automatically.

"Important components of the core include mechanisms for enabling Wireless Debugging and completing the pairing process," ThreatFabric explained. "The core can open the system settings screen, find the relevant menu, enable Developer options, and then enable Wireless Debugging." "When this mode is enabled, it also intercepts the PIN, which is then used to establish a TLS connection to the ADB daemon."

Bazuka supports 23 modules that enable comprehensive data and cryptocurrency theft 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.

The malware has not been attributed to any known threat actor or group. However, the panel has been found to display UI elements in a mix of English and Russian.

ArcticRAT Android Malware Uses LLMs to Extract High-Value Data

Bazuka is not the only Android-focused threat to integrate LLM. Last week, Zimperium zLabs research team detailed another sophisticated and rapidly evolving Android malware called ArcticRAT that supports 129 distinct commands and incorporates LLM APIs into its data exfiltration pipeline to extract high-value financial intelligence from stolen SMS messages and app data.

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in Malware & Botnets

BADBOX 3.0 Botnet Achieves Record 40 Billion Fraudulent Ad Requests Daily
ROMULUS ClickFix Attacks Deliver CNCMachineRMS RAT Through DLL Sideloading Chains
Retadup Worm Resurfaces After Seven Years to Distribute Updated CoffeeLoader Malware