A newly analyzed Android trojan called RatHat hands the hard part of a phone intrusion to a live AI assistant, letting it decide where to tap and scroll instead of following a fixed script. Researchers at Zimperium zLabs documented the malware, which Malwarebytes for Android detects as Android/Trojan.Exploit.RatHat. Because the attack path shifts from device to device, the usual signature- and rule-based mobile defenses have little to match against.
The campaign reaches victims through smishing texts and malicious ads, which lead to fake download pages dressed up in some cases as a streaming app or as a browser such as Chrome. Those pages push people into sideloading a malicious APK, and the app then pressures them to switch on the Android Accessibility Service, using either a made-up network restriction or a bogus financial reward as the excuse. Once accessibility access is granted, the malware can tap through Developer Options on its own.
Read more: Gigabud Android Trojan Clones Banking Apps Into Separate Work Profile
From there RatHat abuses Android Debug Bridge, a legitimate developer tool, and turns on Wireless Debugging to break out of the normal app sandbox. It reads the six-digit pairing code straight off the screen and pairs with the phone by itself, with no helper and no computer involved. That self-pairing produces a shell-level ADB session, and a reverse-proxy client opens a tunnel to the attacker's server that slips past firewalls and NAT.
The payoff is aimed at money. RatHat lays overlays over selected apps, mostly financial ones, to grab sign-in credentials along with one-time passwords and multi-factor authentication codes. It also pulls raw touch coordinates from the input driver and lines them up against known keypad and pattern-lock layouts, which lets it rebuild PIN codes and unlock patterns while sidestepping screen-reading protections. Text messages are intercepted as well.
Removal is the catch. A hidden background program puts the malicious app back after a user uninstalls it, so an infected phone needs a factory reset. The AI service gives the operators semi-autonomous control of the device, according to the analysis.
The findings land in a busy stretch for Android banking threats. Earlier this month Malwarebytes described Gigabud, a banking trojan that builds a separate work profile on the phone and runs a cloned banking app inside it. In February, researchers reported the first Android malware to lean on generative AI for persistence, a family called PromptSpy that sent natural-language prompts with XML screen data to Google Gemini.
PromptSpy, also tracked as MorganArg, appeared built to mimic JPMorgan Chase Bank and was spread through its own website rather than the Play Store, with ESET researcher Lukáš Štefanko saying the campaign looked financially motivated. He noted that "leveraging generative AI enables threat actors to adapt to more or less any device, layout, or operation system version, which can greatly increase the pool of potential victims".
Malwarebytes' guidance is to stick to Google Play or another trusted official store, treat any app that asks for Accessibility Service with suspicion, and leave Developer Options and Wireless Debugging switched off unless there is a real need for them. Android's Advanced Protection Mode limits which apps can request accessibility permissions, and a current real-time anti-malware tool is recommended as a backstop.
