Android Malware Exploits Gemini AI for Runtime Persistence and Stealthy Operations

Researchers Discover Novel Android Malware Leveraging Generative AI

Researchers have discovered a novel Android malware that leverages generative AI to achieve persistence on compromised devices. Dubbed PromptSpy, the malware employs a VNC module to enable its operators to remotely view and control the infected device. Additionally, PromptSpy can collect device information, capture lockscreen credentials, record screen activity to obtain the device’s unlock pattern, and take screenshots.

Unique Approach to Persistence

To maintain persistence, PromptSpy uses a unique approach that involves interacting with Google’s Gemini gen-AI chatbot at runtime. The malware sends a prompt to Gemini, along with an XML file containing information about the device’s UI elements, including their type, text, and position. Gemini responds with JSON instructions that inform PromptSpy where to tap or swipe on the screen to add the malware to the list of recent apps.

Exploiting Accessibility Services

By doing so, PromptSpy ensures its persistence across device reboots. The malware also exploits Accessibility Services to prevent uninstallation, overlaying invisible elements on the screen to block the user from removing it. The only way to uninstall PromptSpy is to reboot the device into Safe Mode, where third-party apps are disabled.

Attribution and Distribution

Researchers have not observed PromptSpy infections in the wild, and it is possible that the malware is a proof-of-concept. However, a domain appears to be designed to deliver the malware to users in Argentina. Analysis suggests that PromptSpy was created by Chinese developers, although this attribution is made with medium confidence, and no link to a specific threat actor has been established.

About Author

Vaibhav

See author's posts