A New Android Malware Kit Lets Crooks Ask an AI Which Victims Are Worth Robbing and Hides Its Work Behind a Fake Home Screen, as Banking Trojans Hand Stolen Texts to Chatbots
Mobile banking fraud just picked up a research assistant. Dutch mobile security firm ThreatFabric has published a teardown of Bazuka, a previously unknown Android malware framework whose control panel includes a built-in chat with a large language model that helps operators decide which infected phones are worth their time. The report, published by ThreatFabric and covered by The Hacker News on November 20, lands days after two other research teams documented Android trojans that pipe stolen text messages into AI models to find bank balances and one-time codes.
For businesses, the message is uncomfortable but clear: the phone in every employee’s and customer’s pocket is now the target of criminal operations that are organized, modular and starting to automate the work that used to need a human fraudster.
What ThreatFabric Found
According to ThreatFabric’s report, its Mobile Threat Intelligence team stumbled onto Bazuka in late October 2026 while investigating what looked like a routine NFC-relay banking trojan, the kind of app that reads a victim’s contactless card and relays it to a criminal elsewhere. Instead of a simple wrapper around open-source code, researchers found a full framework: a dropper or downloader, a core component and 23 separate modules covering screen overlays, keylogging, interception of SMS and two-factor codes, remote control of the device and swapping of cryptocurrency wallet addresses.
The modules read like a fraud toolkit’s table of contents. ThreatFabric lists a clipboard monitor that replaces copied Bitcoin, Ethereum, TRC20 and Solana addresses with the attacker’s own; a keylogger that uses pattern matching to pick out card numbers, wallet addresses and passwords; a module that captures the lock-screen PIN, password or unlock pattern; modules that send USSD codes and manage call forwarding; a reverse proxy that turns the phone into a relay for the criminals’ own traffic; and an “uninstall shield” that fights attempts to remove or force-stop the app.
The Fake Home Screen Trick
One technique was new even to ThreatFabric’s analysts. Bazuka can take a screenshot of the victim’s home screen or lock screen and, on command, display that picture over everything else. To the owner, the phone looks idle and normal. Underneath, operators can be driving banking apps, approving permissions or reading messages. The Hacker News summary describes it as a way to cover up malicious activity happening in the background.
The core also goes after Android’s own defenses. ThreatFabric says it can navigate the settings menus to switch on Developer options and Wireless Debugging, intercept the pairing PIN, and then connect to the phone’s debugging service to grant itself permissions through shell commands. It can also use Accessibility access to disable the Google Play protection component. The firm warns that this kind of “ADB bridging” is becoming standard in banking trojans and could eventually let malware drop modules that survive even after the original app is deleted.
An AI Assistant in the Crime Panel
The detail drawing the most attention is in the operators’ web panel. ThreatFabric found that it supports both locally hosted language models and models reached through an API, and that one sample prompt asks about a victim’s “score.” Combined with the device report each new infection sends home, including a threat-level rating, the researchers conclude that operators can size up how attractive a victim is within seconds and let AI agents help manage infections at scale.
The panel also lets the operators create accounts for “traffers,” the affiliates paid to lure victims, and build droppers with minimal declared permissions, which ThreatFabric says looks designed to help an app slip through Google Play review. The firm stressed it has seen no sign that this has actually happened yet. The lure it did document was an adult-themed “premium TikTok” app. Bazuka has not been attributed to a known group, though the panel mixes English and Russian and one of its servers was hosted in Moscow, according to the report.
Not an Isolated Case
Bazuka arrives in a crowded month. Zimperium’s zLabs team recently detailed ArcticRAT, an Android remote access trojan that The Hacker News reports supports 129 distinct commands and began as spyware aimed at users of 11 cryptocurrency apps. Its newer versions send harvested SMS messages and app lists to an LLM with prompts that cast the model as a professional analyst hunting for financial, banking and trading data. ArcticRAT also abuses Wireless Debugging, uses VPN privileges to null-route network traffic and blind Google Play Protect during installation, and paints a black, un-dismissible screen while attackers work so the device looks switched off.
Earlier in November, STAR Labs documented “Groklem,” an SMS-stealing banker whose code signals point to India and Bangladesh. Its builds are designed to forward a victim’s recent texts to a commercial AI model, using an API key the operator supplies remotely, and get back a summary of bank accounts and balances. STAR Labs traced six builds with six different command servers back to a single Telegram chat, which it read as one operator running a production line. The researchers note they analyzed the code statically and did not observe a key in use.
Why It Matters for Business
SMS codes are getting weaker by the month. All three families read text messages and pull out verification codes. Any company still using SMS as the main second factor for staff logins, payment approvals or customer accounts should treat that as a known weak spot and push toward app-based authenticators, passkeys or hardware keys.
AI lowers the cost of fraud at scale. Sorting thousands of infected phones to find the handful with money in reach used to take human time. If a model does the triage, smaller crews can run bigger operations, and banks, fintechs and crypto platforms should expect more account-takeover attempts that look carefully chosen rather than random.
BYOD policies need a refresh. Bazuka’s ability to enable debugging, capture the lock-screen PIN and turn a phone into a network proxy means a single infected personal device can become a foothold into corporate email, chat and VPN sessions. Mobile threat defense, blocking sideloaded apps on managed devices, and alerts when Developer options or Wireless Debugging are switched on are practical controls.
Customer education still pays. The documented lures are ordinary: adult content, IPTV, fake Google Play pages and lookalike brand apps. Reminding customers never to install banking-adjacent apps from links, and to double-check wallet addresses after pasting, costs little and blunts several of these tricks at once.
The Bottom Line
Bazuka shows mobile malware maturing into a product: modular, sold to affiliates, hardened against removal, and now paired with AI that tells operators where the money is. Organizations that treat phones as a side issue in their security plans are running out of time to change that view.
Sources: ThreatFabric, The Hacker News, Zimperium zLabs, STAR Labs.
Don’t let fraud fears cost you real customers. When worried clients call about a suspicious text or a locked account, MAJ.COM Voice AI answers every inbound call 24/7 and handles outbound sales and customer support, so your business never misses a lead or leaves a customer waiting.
Edited by Coron @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous
Discover more from QUE.com
Subscribe to get the latest posts sent to your email.
