RatHat: New Android Malware Uses a Live AI Assistant to Steal Bank Logins and PINs
Zimperium's zLabs found a banking trojan that abuses Accessibility and Wireless Debugging, lets an AI drive the phone, and reinstalls itself after removal.
Security researchers have documented a new Android banking trojan called RatHat that hands the hard parts of a phone takeover to a live AI assistant. Zimperium's zLabs analysed the malware, and Malwarebytes detects it as Android/Trojan.Exploit.RatHat. It steals banking logins, intercepts one-time passwords, reconstructs screen-lock PINs and patterns, and can survive a normal uninstall (Malwarebytes, Cyber Daily).
How an infection starts
The campaign begins with deception, not a software flaw. Victims receive SMS phishing messages or see malicious ads, forum posts or fake download pages, sometimes dressed up as a streaming app or as Chrome, which push them to sideload an APK from outside an official store. Once installed, the app pressures the user to switch on Android's Accessibility Service, using a made-up network restriction or a bogus financial reward as the excuse.
That permission is the key. It lets the app read the screen and tap on the user's behalf. RatHat uses it to open Developer Options, enable Wireless Debugging, read the six-digit pairing code and pair with the phone's own Android Debug Bridge, with no computer involved. That gives the attacker a shell-level session that gets around Android's normal app sandbox.
| Stage | What RatHat does |
|---|---|
| Lure | Smishing, ads, forums or fake download portals push a sideloaded APK |
| Permission | Pressures the user to enable Accessibility Service |
| Escalation | Enables Wireless Debugging and pairs with on-device ADB for shell access |
| Theft | Overlays on banking, crypto, payment and messaging apps; SMS interception; screen capture; touch-coordinate recording |
| Persistence | Hidden process reinstalls the app and restores permissions after removal |
Why the AI part matters
Most mobile malware follows a fixed script. RatHat gives a live AI assistant access to the Android accessibility tree, so it can interpret the screen and decide where to tap, scroll or type. Because the attack path changes from device to device, signature- and rule-based defences have less to match against. Zimperium says the AI prompts in the code suggest a Chinese-speaking threat actor.
Two caveats from analysts: shell access is serious but is not the same as unrestricted root access on every Android model, and RatHat is not a remote-infection tool on its own, since the user must still sideload it and grant permissions (Gridinsoft).
What to do if you think you are infected
Act quickly. Change banking passwords from a separate, trusted device, contact your bank to review access and recent transactions, and consider a factory reset before restoring only essential, trusted data, since removing the app alone may not be enough.
To avoid it, install apps only from official stores, never install apps from links in text messages or ads, be wary of any app asking for Accessibility Service or Device Admin permission for reasons unrelated to accessibility, and do not enable Developer Options or Wireless Debugging unless you know why you need them. Android's Advanced Protection Mode, which restricts which apps can request accessibility permissions, adds a further layer.
For Indian users
The lures RatHat relies on, sideloaded APKs and fake reward or update messages, are common in India, where mobile banking and UPI transactions are protected by one-time passwords that this malware is designed to intercept. The published research does not name Indian banks or UPI apps as confirmed targets, but the overlay technique works against any app.
What to watch next
Watch whether Google tightens Accessibility Service permissions for sideloaded apps, whether on-device protections such as Samsung's Scam Detection catch this style of attack, and further research on how the AI component behaves in the wild.
