AI‑Powered RatHat Android Malware Automates Device Control and Harvests Credentials
What Happened — Researchers at Zimperium zLabs identified a new Android malware family, RatHat, that leverages a large‑language‑model‑driven UI‑automation engine to navigate compromised phones. The malware abuses Accessibility permissions, enables Developer Options and Wireless Debugging, and uses ADB‑level agents to maintain persistence and exfiltrate banking, cryptocurrency and OTP data.
Why It Matters for Trust & Control Assurance
- Demonstrates how privileged Android controls (Accessibility, Developer Options, ADB) can be subverted, underscoring the need for continuous monitoring of high‑risk permissions.
- AI‑driven automation makes malicious behavior harder to detect with static signatures, highlighting the importance of evidence‑based control assurance that captures behavioral anomalies.
- Provides a concrete scenario where a control‑assurance program must prove that privileged access is limited, logged, and regularly reviewed.
Who Is Affected — Mobile‑first financial services, cryptocurrency wallets, telecom carriers, and any organization that distributes or relies on Android apps for customer interaction.
Recommended Actions
- Enforce strict app vetting and block installations from unknown sources.
- Restrict Accessibility and Developer Options to a whitelist of approved apps; monitor ADB activation events.
- Deploy behavioral analytics that flag AI‑generated UI interactions or anomalous accessibility usage.
- Incorporate AI‑tool usage policies into your mobile security governance framework.
Source: BleepingComputer
Technical Notes — RatHat is delivered via malvertising, SMS phishing, and rogue APK sites. It serializes the Android Accessibility tree to XML, sends it to an external AI assistant for navigation commands, and uses two native agents (liblocal-service.so, libmedia_codec.so) for persistence and reverse‑proxy tunneling. The malware captures UI overlays for banking/crypto apps, intercepts SMS OTPs, and can restore itself after removal attempts. Source: same