RatHat Android Trojan Integrates Generative AI for Real-Time Accessibility Tree Navigation

The dropper carries two encrypted assets that install a malicious app, Go-based C&C agent, and FRP-derived reverse proxy. The app requests Device Admin rights, overlays WebView phishing pages for banking apps, intercepts SMS, and uses getevent at the kernel level to reconstruct PINs from raw touch coordinates. A hidden background service reinstalls the package if removed and maintains ADB access for persistence. Zimperium's report supplies the primary technical artifacts: AI prompts written in Chinese that instruct the model on layout parsing and action sequences. This matches procurement patterns where Chinese-language tooling appears in both state-linked and criminal Android campaigns, distinct from the Russian or Brazilian banking trojan clusters. No independent sample analysis or C2 infrastructure mapping has yet confirmed actor identity beyond the prompt language. Prior coverage treated the AI component as a novelty rather than an operational multiplier. The real shift is automation of privilege escalation and anti-uninstall logic, lowering the skill floor for sustained remote access. Similar patterns appear in North Korean Linux toolkits that also chain accessibility abuse with proxy tunnels. Expect variant droppers leveraging open-source LLMs to appear within six months, targeting the same Accessibility Service vector on devices running Android 12-14.