“Bazuka” Android Malware Uses Overlay Trick, LLMs to Steal Credentials

A new Android malware framework dubbed “Bazuka” is raising alarm among cybersecurity experts for its highly modular design, integration of large language models (LLMs), and ingenious use of screen overlays to mask its activities. Discovered by mobile security firm ThreatFabric in late October 2026, Bazuka features a dropper/downloader component, a core module, and 23 plugins that support everything from credential theft to remote device control.

What is Bazuka and how does it work?

Bazuka begins its attack with a dropper or downloader disguised as legitimate apps. One fake app mimics a “premium” version of TikTok under the package name com.tiktok.premium. The dropper requests permissions to install apps from third-party sources, then either fetches a standalone APK (like com.tiktok.premium.pro) or downloads the main payload directly. Once active, Bazuka’s core orchestrates everything: loading additional modules, handling command-and-control (C2) communication, collecting device info, and managing permissions through Android’s accessibility services and ADB (Android Debug Bridge).

A standout tactic: Bazuka can take a screenshot of the current home or lock screen and overlay it at opportune moments. This faux interface hides suspicious behavior running in the background—anything from permission dialogs to fraud-inducing overlays. In parallel, modules monitor keystrokes, intercept 2FA codes, watch the clipboard for crypto-wallet addresses, and siphon device location, contacts, SMS, camera/microphone data.

LLMs and the threat actor toolkit

Part of what makes Bazuka particularly dangerous is its AI-powered control panel. The framework offers operators tools to score victims for prioritization and manage infection campaigns using LLMs, whether hosted locally or accessed via API. This includes building customized droppers that adjust Android manifest permissions—potentially as a tactic to slip past app store reviews.

The 23 modules, each specialized, let attackers, for instance, swap clipboard wallet addresses, gain shell access via ADB, manage files, spy on communications, disable uninstall attempts, or even toast system notifications to mislead users. ADB abuse is another core feature: Bazuka can enable wireless debugging, navigate settings, intercept PINs, and establish secure connections to device components. The malware also has a self-destruct mechanism to wipe traces if its operations are threatened.

Bigger picture & related threats

Bazuka isn’t the only Android malware to embrace AI. ArcticRAT, another recently analyzed Trojan, leverages LLM APIs to sift through stolen SMS messages and app data for banking and crypto intelligence. ArcticRAT also uses “black screen” overlays that make devices appear locked while malware runs in the background.

No threat actor has claimed responsibility for Bazuka, but its mix of English and Russian interface elements hints at possibly Eastern European origins. Analysts are particularly concerned about its use of accessibility services and ADB—tools typically managed by advanced financial Trojans—to retain privileges and evade detection.

For Android users, Bazuka marks a significant escalation in both tactics and tooling. The overlay screenshot trick is novel, giving attackers a stealthy way to hide what’s really happening on the device. Coupled with AI to prioritize victims, this malware framework might push the next generation of mobile threats into territory that was once the domain of desktop or server attacks.

What to watch:how app stores respond to apps abusing manifest permissions; whether wireless debugging abuse becomes more widespread; and the rise of victim-scoring via LLMs—if this becomes standard, security tools and mobile OS defenses will need to evolve fast.